<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Tim Stolte on Medium]]></title>
        <description><![CDATA[Stories by Tim Stolte on Medium]]></description>
        <link>https://medium.com/@stol_?source=rss-865ada47c174------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*WXuv2qQHxuVHmo6nW-RcEg.jpeg</url>
            <title>Stories by Tim Stolte on Medium</title>
            <link>https://medium.com/@stol_?source=rss-865ada47c174------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 06:38:57 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@stol_/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Lump-sum or Dollar Cost Averaging for bitcoin?]]></title>
            <link>https://medium.com/amdax-asset-management/lump-sum-or-dollar-cost-averaging-42c8f5bb9938?source=rss-865ada47c174------2</link>
            <guid isPermaLink="false">https://medium.com/p/42c8f5bb9938</guid>
            <category><![CDATA[investing]]></category>
            <category><![CDATA[bitcoin]]></category>
            <category><![CDATA[dollar-cost-averaging]]></category>
            <category><![CDATA[quantitative-research]]></category>
            <category><![CDATA[lump-sum-investment]]></category>
            <dc:creator><![CDATA[Tim Stolte]]></dc:creator>
            <pubDate>Tue, 18 Apr 2023 14:52:49 GMT</pubDate>
            <atom:updated>2023-04-18T14:53:58.444Z</atom:updated>
            <content:encoded><![CDATA[<p>Suppose you have a significant amount of money to invest in an asset of choice and you are not sure where the market might be heading in the near future. Should you just invest it all and hope for the best? Or is it better to gradually build your portfolio? Both options have their pros and cons, and choosing the right one can significantly impact your investing experience.</p><p>Perhaps the most common strategy is called either lump-sum or buy-and-hold, which, as its name suggests, involves deploying all your available capital at once and holding on to that position. It is a good way to reach a certain financial goal in the long run, but it does introduce a huge risk to a portfolio. Retail investors are notoriously ‘<em>late to the party</em>’. They often decide to take on exposure to an asset once it has already experienced significant upside. This increases the chances of buying a market top, resulting in a portfolio that quickly decreases in value, only turning a profit months or even years after the initial investment.</p><p>To solve the timing risk problem, some investors turn to another strategy: Dollar Cost Averaging, often referred to by the abbreviation DCA. This strategy calls for investing a fixed part of their savings with a predetermined frequency. It usually boils down to a monthly investment, but (bi-)weekly intervals are also common.</p><p>By definition, DCA decreases the speed at which an investor gains market exposure. So with DCA, investors avoid losses in downward moving markets as they have not yet invested all of their capital. But risk mitigation goes both ways. DCA investors miss out on upside in upward moving markets. This ultimately raises one crucial question: does the risk management in DCA outweigh its potential lack of returns?</p><h3><strong>A case study: bitcoin</strong></h3><p>In order to answer our research question, we turn our attention to one asset in particular: bitcoin. The logarithmically scaled price chart below clearly shows at which points in time a lump-sum strategy would have been a poor choice. Entering a position in late 2013, 2017 or 2021 would have resulted in a decrease in portfolio value of around 80%. And as it happens, these are exactly the times at which retail market participation is at its peak, often forming speculative bubbles.</p><p>Of course, this is only one side of the coin. Buying bitcoin during market bottoms in 2012, 2015 or 2019 would have resulted in huge upside for lump-sum, while DCA would have limited these profits by buying at higher prices after the market already bottomed out.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*duqlz4nMlRmGDmpqekwBCQ.png" /></figure><p>There is clearly a trade-off between the cautiousness of DCA and the return potential of lump-sum. But which of the two strategies is closest to the perfect balance? Before we jump to data analysis, we must note that the optimal combination depends on each individual’s risk appetite. If one is not bothered by any portfolio fluctuations and has a long enough investment horizon, lump-sum would have been optimal most of the time. But being caught off-guard by huge price drops can have massive repercussions if one is forced to exit a position <strong>before</strong> the market turns in their favour.</p><p>In our research, we will use two well-known and simple measures to gauge the effectiveness of DCA on risk management and how decreased risk relates to overall portfolio performance:</p><ol><li><strong><em>Maximum drawdown</em>:</strong> the largest observed drop that a portfolio experienced in a given time frame. It yields an approximation for the worst-case scenario for that period. For instance, if the maximum value of bitcoin was $40k in 2022, and decreases to a minimum value of $20k <strong>afterwards</strong>, the <em>maximum drawdown</em> in that year is -50%.</li><li><strong><em>Calmar ratio</em></strong>: the average portfolio return relative to the <em>maximum drawdown. </em>If one portfolio has a higher <em>Calmar ratio</em> than another portfolio, it means that it has a better risk-return performance.</li></ol><p>Let’s apply some backtesting and find out how the lump-sum and both monthly and weekly DCA strategies have performed in the past ten years. The backtest details and results summary can be found in the Appendix.</p><h3>Results</h3><p>We start off with the most straightforward performance metric: the actual returns per year for each strategy. These returns are measured from the start to the end of the year and give us insights into whether it is profitable to deploy cash all at once (lump-sum) or gradually (DCA).</p><p>In the figure below, lump-sum generally shows more extreme returns, both in positive and negative directions. Especially the returns in 2013 are exceptional (lump-sum returns are over 5,400% and are not fully included in the plot for the sake of readability). The average annual returns are over three times higher than those of the DCA strategies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4imdzIGiamd0QmqSzZ4MEA.png" /></figure><p>We know that returns do not paint a strategy’s full picture, so we also need to consider risk measures such as the <em>maximum drawdown</em>. The figure below shows the results for each individual year since 2012. We notice that lump-sum has larger drawdowns and is consistently riskier than DCA. Especially in years with extreme crashes (like COVID in March 2020), the difference in drawdown is quite substantial. On average, the lump-sum portfolio drops around 20% more from top to bottom. And between the two DCA strategies, the weekly DCA appears to be the least risky.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L2u7HcImNzSaaDMV9jrLuQ.png" /></figure><p>But what does the limited downside risk of DCA tell us about the overall portfolio performance? Putting the returns and <em>maximum drawdowns </em>into a single equation gives us the <em>Calmar ratio</em>. This ratio shows very large fluctuations across all the years in our sample period. Sometimes the DCA strategies outperform lump-sum and sometimes it is the other way around. But on average, the <em>Calmar ratio</em> is practically identical for all three strategies. In other words, all strategies show similar performance if we adjust for the level of risk.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pEws3LdlCW0fQ599h184Ww.png" /></figure><h3>Conclusion</h3><p>In this article, we analysed the performance of two retail investment strategies: lump-sum and Dollar Cost Averaging (DCA). The latter mitigates the timing risk of the former, but also potentially limits profits in an upward moving market. The results show that DCA consistently lowers <em>maximum drawdowns</em> in bitcoin-only portfolios, while keeping the overall performance equal to that of a lump-sum strategy. This leads to the conclusion that DCA is a suitable solution for careful investors that seek to manage the risk in their portfolio while still gaining exposure to a volatile asset class such as crypto.</p><p>In the end, investors should consider their risk tolerance and investment goals before choosing what strategy they are most comfortable with. But the effort to buy an safely store bitcoin on a regular basis can also play an important role in this decision. At Amdax, we try to lower the threshold to opt for a DCA strategy by providing a ready-to-use <a href="https://www.amdax.com/en/products-services/products-services-2-2/dca-periodic-crypto-purchases">savings plan</a>.</p><h3>Appendix</h3><h4>Backtest details</h4><ul><li>The total amount of investable capital is assumed equal for all strategies.</li><li>We assume a fixed transaction fee rate at 0.7% per transaction and fixed custody fee at 0.6% per year.</li><li>We assume an annual risk-free rate equal to the <a href="https://fred.stlouisfed.org/series/RIFSGFSW04NA">4-week T-bill rate</a>.</li></ul><h4>Results summary</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/740/1*G2p1-82SysvP1LIuUQDhbQ.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=42c8f5bb9938" width="1" height="1" alt=""><hr><p><a href="https://medium.com/amdax-asset-management/lump-sum-or-dollar-cost-averaging-42c8f5bb9938">Lump-sum or Dollar Cost Averaging for bitcoin?</a> was originally published in <a href="https://medium.com/amdax-asset-management">Amdax Asset Management</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Bitcoin Hash Rate: a Failing Investment Tool]]></title>
            <link>https://medium.com/amdax-asset-management/the-bitcoin-hash-rate-a-failing-investment-tool-e343cd99719c?source=rss-865ada47c174------2</link>
            <guid isPermaLink="false">https://medium.com/p/e343cd99719c</guid>
            <category><![CDATA[quantitative-research]]></category>
            <category><![CDATA[backtesting]]></category>
            <category><![CDATA[hash-rate]]></category>
            <category><![CDATA[investing]]></category>
            <category><![CDATA[bitcoin]]></category>
            <dc:creator><![CDATA[Tim Stolte]]></dc:creator>
            <pubDate>Fri, 21 Oct 2022 12:20:15 GMT</pubDate>
            <atom:updated>2022-10-21T12:33:16.754Z</atom:updated>
            <content:encoded><![CDATA[<h4>Part II: The Backtest</h4><p>In <a href="https://medium.com/amdax-asset-management/the-bitcoin-hash-rate-a-failing-investment-tool-725a12d7f392">Part I</a>, we have formed the hypothesis that the Hash Ribbon does not yield a reliable strategy in recent years. But our findings have not yet been supported by hard data. In this article, we turn to backtesting the strategy in order to properly gauge its performance.</p><h3>A proper benchmark</h3><p>The Hash Ribbon is a metric that signals buying opportunities. There are no sell signals in the strategy. So in the construction of our backtest, we have to be aware that we cannot exit a Bitcoin (BTC) position, we can only enter.</p><p>The first thing to consider is the benchmark. Or in other words, what do we compare the strategy’s performance with? We could, of course, simply calculate the USD returns of the Hash Ribbon strategy, effectively using the value of the dollar as a benchmark. But since the USD value of Bitcoin itself has already increased around 100,000% in the last 12 years, we would be calculating the performance of BTC rather than the performance of the strategy.</p><p>To correct for this, we could calculate the strategy’s returns in terms of BTC. This means that we use a buy-and-hold strategy as a benchmark. The problem with that is that buy-and-hold immediately deploys all available capital at the start, while the Hash Ribbon strategy can only deploy a chunk of the available capital every time a buy signal flashes. With Bitcoin’s historical price path, deploying capital early almost always leads to higher returns. This again results in an unfair advantage for the benchmark strategy.</p><p>The most objective way to assess the performance of this strategy is to compare it with a monthly dollar-cost averaging (DCA) strategy. This is a buying strategy that converts a fixed amount of USD to BTC every month.</p><h3>Methodology</h3><p>To properly compare the Hash Ribbon and DCA strategies, we assume that every month 1,000 USD will become available to convert to BTC. The DCA strategy converts this monthly amount immediately, but the Hash Ribbon approach is a little different. When the monthly 1,000 USD becomes available, either:</p><ul><li>no buy signal flashes this month, so we save the 1,000 USD to deploy when the next signal flashes. We can thus build up available USD capital over multiple months. Or…</li><li>a buy signal flashes somewhere this month, so on that day we convert the 1,000 USD together with any potentially saved up USD capital from previous month(s).</li></ul><p>This approach ensures that both strategies use the same amount of USD capital, which can potentially be deployed at the same time. So in this theoretical set-up, no strategy has a definite advantage over the other.</p><h3>Performance results</h3><p>Below we have the portfolio value in USD of the Hash Ribbon (in blue) and DCA strategy (in red) from 27 December 2011 until 30 September 2022. We choose this starting date because it is the first Hash Ribbon buy signal, so both strategies start off simultaneously with an immediate deployment of the first monthly 1,000 USD.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oogApKYzgUrkBoU-OMkIHg.png" /></figure><p>To simplify comparing the two strategies, we divide the Hash Ribbon portfolio value by the DCA portfolio value. This yields the following series, where 1 represents the DCA performance. A value higher than 1 indicates that the Hash Ribbon signal outperforms DCA and the opposite holds for a value below 1. We find that the Hash Ribbon performs 21% worse than DCA over this full period. However, this conclusion comes with a few snags.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7fXguqMrJF0T1ozw8-sn_w.png" /></figure><p>Firstly, note that the relative performance becomes less volatile with each month. This is because our choices at the start affect each choice that comes after it. As we accumulate capital, the fixed monthly 1,000 USD becomes less and less proportional to our total wealth. To be exact, the 1,000 USD in the last month has 130 times less impact on our performance than the first 1,000 USD.</p><p>Secondly, it is very hard to outperform DCA when the price increases for a long period of time. It immediately deploys all capital whenever it becomes available. This is close to the optimal strategy in bull markets, because it ensures that we always buy at one of the lowest points possible. Hence, we find that DCA already massively outperforms the Hash Ribbon strategy in late 2012 and 2013.</p><p>Unfortunately, there is little we can do to fully adjust for these two points, as they are inherent to the nature of buying strategies. But we can try to limit the effects by decreasing the sample period.</p><p>Our methodology remains the same, but we now analyse the performance for multiple starting dates. We start at buy signals that have occurred more recently. An advantage of a more recent sample period is that we can find out whether the strategy would hold up in present times.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2U2X2rwNW1cqfYLJBCUl-g.png" /></figure><p>The Hash Ribbon can be more profitable than DCA during some short periods. But out of the five starting dates, only one yields actual outperformance over the full sample period. On average, the Hash Ribbon performs slightly worse (0.02%) than DCA.</p><h3>An alternative buying strategy</h3><p>Given that our new sample period still contains a bull market, one might argue that DCA is as close to an optimal buying strategy as we can get and, therefore, that it’s unfair to use it as a benchmark. In order to find out whether this is actually the case, let’s consider another very simple buying strategy that may outperform DCA.</p><p>For this strategy we need the realised price. The realised price is a well-known metric that is, just like the hash rate, native to cryptocurrencies. It’s defined as the average price at which every Bitcoin was last moved on the blockchain. It’s also often referred to as the <em>Bitcoin cost basis</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ypt584fPYkVxe1FXvXBaPw.png" /></figure><p>The strategy proposal is as follows: buy when the market (actual) price falls below the realised price. We can hypothesise that this marks periods when Bitcoin is “too cheap”, because its price is lower than what the average holder was willing to pay for it.</p><p>Let’s first take a look at the average returns after these realised price signals and compare those to the average returns after Hash Ribbon signals. Specifically, we consider 14-day, 30-day, 60-day and 90-day returns in three sample periods, starting in 2011, 2015 and 2019. This data should give us some preliminary insights before we start backtesting. We find that both signals are generally followed by above-average returns. But there is little statistical evidence that either one of the strategies yields better returns than the other. In the largest sample, the Hash Ribbon significantly outperforms for 30-day and 60-day returns. However, the results in the most recent subsample are in favour of the realised price strategy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*t8li750qI1_egtv5Qy_W9w.png" /><figcaption>The average 14-day, 30-day, 60-day and 90-day return after all dates, after Hash Ribbon signals and after realised price signals for three sample periods. Statistically significant differences in population medians are calculated using a Kruskal-Wallis test and are denoted with an using a 5% significance level.</figcaption></figure><p>The average returns do not show a clear winner between both strategies, so we assess the realised price strategy performance using the same backtest as before. The figure below displays the BTC price in grey and the relative performance series in the different shades of blue. It shows that we almost always significantly outperform DCA in the long run.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S0uXx4Duwnpn9d1GR7af_w.png" /></figure><p>Of course, we should note that there are considerable drawdowns in this relative performance compared to the ones we have seen for the Hash Ribbon. Thus, our quest for the optimal buying strategy has not yet come to an end. But we do know that it is possible to challenge DCA for this top spot with a strategy as simple as this one.</p><h3>Conclusion</h3><p>While the Hash Ribbon has reasonably identified global price bottoms in the past, it seems that the most recent buy signals fall short of predicting price bottoms. Using careful backtesting methods, we conclude that our strategy based on the Hash Ribbon metric does not reliably outperform a DCA benchmark strategy.</p><p>We find that we can achieve better returns using a very simple strategy based on the realised price. This raises the suspicion that there are a multitude of buy signals that yield outperforming strategies. All in all, we can argue that it’s unlikely that the Hash Ribbon metric contributes to an optimal investment strategy.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e343cd99719c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/amdax-asset-management/the-bitcoin-hash-rate-a-failing-investment-tool-e343cd99719c">The Bitcoin Hash Rate: a Failing Investment Tool</a> was originally published in <a href="https://medium.com/amdax-asset-management">Amdax Asset Management</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Bitcoin Hash Rate: a Failing Investment Tool]]></title>
            <link>https://medium.com/amdax-asset-management/the-bitcoin-hash-rate-a-failing-investment-tool-725a12d7f392?source=rss-865ada47c174------2</link>
            <guid isPermaLink="false">https://medium.com/p/725a12d7f392</guid>
            <category><![CDATA[investing]]></category>
            <category><![CDATA[quantitative-research]]></category>
            <category><![CDATA[hash-rate]]></category>
            <category><![CDATA[bitcoin]]></category>
            <dc:creator><![CDATA[Tim Stolte]]></dc:creator>
            <pubDate>Mon, 17 Oct 2022 14:07:52 GMT</pubDate>
            <atom:updated>2022-10-17T14:07:52.545Z</atom:updated>
            <content:encoded><![CDATA[<h4>Part I: Wen Bottom?</h4><p>At the time of writing, Bitcoin (BTC) is going through a rough bear market. Over the span of 11 months, the price has fallen to less than a third of its all-time high and has yet to show any signs of recovery. Since BTC still holds well over 40% of the crypto market share, most crypto investors are yearning for a trend break in the decreasing BTC price series. The question is when we will find a floor price, after which the price resumes its long-term upward path. Or simply put in true CT (Crypto Twitter) fashion: wen bottom?</p><p>Many crypto analysts are currently occupied with finding the answer to this question. And a multitude of parallels can be drawn with previous bear markets. The main approach consists of three steps:</p><ol><li>Select a metric that may provide useful information on price bottoms.</li><li>Find out what values this metric showed during previous bottoms.</li><li>Wait for the metric to hit those levels and call the bottom.</li></ol><p>Some metrics flashed bottom signals around a BTC price of $40k, others flashed at $20k and some have not flashed at all (yet). Admittedly, we should account for some wiggle room, but there is reason enough to believe that we cannot accurately predict the level and timing of long-term price bottoms with historical data. There is simply too much uncertainty.</p><p>Still, analysts keep coming up with bottom signals. One of them is particularly popularised lately, and it has everything to do with the hash rate.</p><h3>What is the hash rate?</h3><p>Proof-of-Work blockchains like Bitcoin are powered by miners that solve cryptographic problems in exchange for BTC rewards. Due to the nature of cryptography, solving the problems involves computational guesswork rather than complex mathematics. A miner that guesses faster can guess more often, making it more likely to receive rewards instead of other miners. So mining is all about speed.</p><p>The hash rate is the total number of guesses per second on the Bitcoin network. The higher the hash rate, the more miners are actively competing for the reward, making the network more secure, which increases the value of the Bitcoin network. And as such, there is reason to believe that the price increases if the hash rate increases. But does this idea actually hold in practice?</p><h3>Price effects</h3><p>The figure below shows the weekly average of the Bitcoin hash rate against the price. The hash rate generally increases and only experienced a few downturns since 2014. These downturns sometimes coincide with sharp price decreases (end of 2018 and in 2021), but they also happen during price increases (during bull runs in 2017 and 2020). Conversely, long periods with decreasing price trends are not characterised by falling hash rates.</p><p>So while there are many examples of the hash rate and price moving in the same direction, there are just as many examples of them moving in opposite directions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*deBqOxGdQ9Hm0d_OcY-_OA.png" /></figure><p><em>*I took weekly moving averages and excluded data up until 2014 to increase visibility of the hash rate data over the full sample period. These choices are without loss of generality and do not affect the conclusions in any way.</em></p><p>This empirical evidence is also supported by academic literature. Contrary to most other crypto metrics, the hash rate is well-researched by academics. Recent papers conclude that there is no statistically significant causal relationship between hash rate and price¹ and that if anything, price drives hash rate instead of the other way around².</p><h3>The Hash Ribbon</h3><p>Even though the hash rate doesn’t appear to affect price, it is often used by crypto analysts. Introduced by <a href="https://medium.com/u/3e17a788b1e">Charles Edwards</a> in 2019, the Hash Ribbon is a metric that uses the hash rate to find price bottoms³. In particular, it seeks to identify periods in which miners are forced to shut down because of a market contraction, causing the hash rate to drop significantly. It then flashes a buy signal when the hash rate stops decreasing and the price starts increasing, indicating the end of the miner capitulation and the start of a potential price surge. For a more detailed description, I refer to his <a href="https://medium.com/capriole/hash-ribbons-bitcoin-bottoms-60da13095836">introductory article</a>.</p><p>On first glance, the Hash Ribbon buy signals look impressive. They generally pinpoint buying opportunities followed by large upside price movements and limited drawdowns. However, one might argue that those opportunities are not rare in a price chart that looks like Bitcoin’s.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sdcNnzrc1Rx16YGKF_MmWQ.png" /></figure><p>Based on the most recent signals, it is safe to say that the performance of the Hash Ribbon has deteriorated a bit. The subsequent price increases are not as significant as they could be and some are even followed by considerable drawdowns.</p><p>The Hash Ribbon is based on <a href="https://www.investopedia.com/terms/c/crossover.asp">cross-overs</a> of different moving averages with varying window sizes. These window sizes are the core parameters of the metric and are chosen rather arbitrarily. But, as the original article states, the choices are not overly important and the results remain similar if other window sizes are selected. While that may hold for most older signals, the claim should be withdrawn for recent signals.</p><p>Let’s zoom in on the hash rate cross-over. The original metric uses a cross-over of the 30-day and 60-day moving average of the hash rate. We find that, if we make small adjustments to the latter, we end up with some very poor buy signals. As shown in the figure below, setting the window size of the second moving average to anywhere between 73 and 90 days causes the buy signal to flash at $57.4k in May 2021, right before the drop to $30k. These parameters are, of course, just as arbitrary as the original 60 days. We can make more adjustments that lead to similar conclusions, but this one example already proves that the Hash Ribbon performance is anything but stable.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vgeoMrg8hm5v5G2JotRFUg.png" /></figure><h3>Conclusion</h3><p>So far, I have empirically shown that the value of the hash rate as an investment tool is often overstated. It can explain developments in the Bitcoin mining industry, but there is no evidence of any causal relationship from the hash rate to the price. This leads to doubts on the validity of investment signals such as the Hash Ribbon. In Part II of this article, I provide quantitative substance on the subject and provide proof that the Hash Ribbon strategy indeed fails to outperform simple benchmark strategies.</p><h3>References</h3><p>[1] Shanaev, S., Sharma, S., Shuraeva, A., &amp; Ghimire, B. (2019). The marginal cost of mining, Metcalfe’s law and cryptocurrency value formation: Causal inferences from the instrumental variable approach. <em>Metcalfe’s Law and Cryptocurrency Value Formation: Causal Inferences from the Instrumental Variable Approach (June 7, 2019)</em>.</p><p>[2] Fantazzini, D., &amp; Kolodin, N. (2020). Does the hashrate affect the bitcoin price? <em>Journal of Risk and Financial Management</em>, <em>13</em>(11), 263.</p><p>[3] Edwards, C. (2019). Hash Ribbons &amp; Bitcoin Bottoms. <a href="https://medium.com/capriole/hash-ribbons-bitcoin-bottoms-60da13095836">https://medium.com/capriole/hash-ribbons-bitcoin-bottoms-60da13095836</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=725a12d7f392" width="1" height="1" alt=""><hr><p><a href="https://medium.com/amdax-asset-management/the-bitcoin-hash-rate-a-failing-investment-tool-725a12d7f392">The Bitcoin Hash Rate: a Failing Investment Tool</a> was originally published in <a href="https://medium.com/amdax-asset-management">Amdax Asset Management</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Bitcoin’s power-law corridor debunked]]></title>
            <link>https://medium.com/amdax-asset-management/bitcoins-power-law-corridor-debunked-1b40783657bf?source=rss-865ada47c174------2</link>
            <guid isPermaLink="false">https://medium.com/p/1b40783657bf</guid>
            <category><![CDATA[power-law]]></category>
            <category><![CDATA[quantitative-research]]></category>
            <category><![CDATA[bitcoin]]></category>
            <category><![CDATA[econometrics]]></category>
            <category><![CDATA[quantitative-finance]]></category>
            <dc:creator><![CDATA[Tim Stolte]]></dc:creator>
            <pubDate>Fri, 02 Sep 2022 13:39:15 GMT</pubDate>
            <atom:updated>2022-09-02T13:39:15.765Z</atom:updated>
            <content:encoded><![CDATA[<p>Three years ago in 2019, long-term Bitcoin price models popped up one after another. They all attempted to forecast Bitcoin prices a few years into the future. Some failed, but some happened to align with the price dynamics that came after it. The most famous one remains the <a href="https://medium.com/@100trillionUSD/modeling-bitcoins-value-with-scarcity-91fa0fc03e25">Stock-to-Flow model</a> by <a href="https://medium.com/u/bcb63a182704">PlanB</a>. It received a lot of criticism over time, and rightly so. Check out my <a href="https://medium.com/amdax-asset-management/stock-to-flow-a-philosophy-not-a-model-7d22c2eb7fd0">previous article</a> to find out what’s exactly wrong with it.</p><p>Another well-known prediction method was put forward a few months later by <a href="https://medium.com/u/8a70b212d583">Harold Christopher Burger</a>. He proposes a so-called <a href="https://medium.com/quantodian-publications/bitcoins-natural-long-term-power-law-corridor-of-growth-649d0e9b3c94">power-law corridor growth</a> model. It simply documents a relationship between the relative change in time and the relative change in Bitcoin price. The model hasn’t been criticised quite as much as Stock-to-Flow, but that doesn’t make it any less flawed.</p><p>In this piece, I’ll provide a deep-dive into the power-law model and attempt to explain why it’s a highly inadequate method to predict the Bitcoin price. My arguments include both logical and statistical reasoning, but do not require any prior knowledge on the matter.</p><h3>Time does not pass exponentially</h3><p>Interpreting the logic behind a model can be hard, but is absolutely crucial in any case. For a relevant example, I refer to my <a href="https://medium.com/amdax-asset-management/all-models-are-wrong-and-this-one-isnt-even-useful-a404fea38ca9">recent article</a> on <a href="https://medium.com/u/bcb63a182704">PlanB</a>’s “quant investing strategy”. In the power-law model, <a href="https://medium.com/u/8a70b212d583">Harold Christopher Burger</a> starts by plotting the Bitcoin price series (on the y-axis) over time (on the x-axis). After observing a highly non-linear relationship between the two, he continues by <a href="https://www.investopedia.com/ask/answers/05/logvslinear.asp">logarithmically scaling</a> <em>both</em> axes. This eventually results in a seemingly linear relationship between the logarithm of price (log-price) and the logarithm of time (log-time):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/952/1*7LJFcJaOS3y-wGu9lIhRqQ.png" /></figure><p>Logarithmically scaling time is possibly the weirdest thing I have ever seen in time series analysis. The whole point of log-scaling is to ensure that change is interpreted in relative terms.</p><p>Let’s imagine the price of an imaginary asset. If the price is not scaled, a change from 2 to 3 is considered equal to a change from 1 to 2, since there is only one unit of change in both cases. But we all know that this yields a distorted picture, because a change from 2 to 3 is much smaller in relative terms (50% vs 100% increase). We thus logarithmically scale the variable such that a change from 2 to 3 is considered twice as small as a change from 1 to 2.</p><p>A scaled time axis, on the other hand, makes no sense at all, because we want the change from 2022 to 2023 to be <em>exactly</em> equal to the change from 2011 to 2012. The amount of time in a year is constant. If we were to log-scale time, we are effectively modelling the real-world time to pass increasingly faster. That is ridiculous.</p><p>So what actually happens if we apply this transformation? It has numerous consequences, but I’ll name three here.</p><h4>Robustness</h4><p>Sample periods matter <em>a lot </em>for this particular method. He applies log-scaling to make the plot resemble a linear relationship. But watch what happens when I adjust the starting date to 2014.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/947/1*Y5oqVjdew3pp17rrNv04Ug.png" /></figure><p>Anyone can see that this relationship is miles away from a linear one. And that is just by simply moving the starting date a little into the future. We thus find that we can show the model’s inability by playing around with just one single parameter. In statistics, good models are little affected by underlying assumptions, which is a property called <em>robustness</em>. This power-law growth model clearly doesn’t belong to those models.</p><h4>Logarithmic dates</h4><p>It’s impossible to actually apply a logarithm to a date. It only works on positive numbers. So in order to create this plot, I converted the dates to positive integers, starting with 1 on the first date, 2 on the second date and so on until the last date. But it appears there’s a problem with that approach.</p><p>In the figure below, I included log-log plots with different starting integers. Note that the straightforward approach (starting with 1) produces nothing like a linear relationship. The same holds for very large starting integers like 1000, which look like common log-plots. He uses an absurdly arbitrary starting integer that equals the number of days since 1 January 2009, which equals 577 in my case (it’s 563 in the original model). There’s no logic or wisdom there, just pure guesswork and picking whatever looks nice...</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/956/1*mtFb5C1XAl-uvrDFyiJ6zA.png" /></figure><h4>Linear approximation</h4><p>He identifies the linear approximation of the log-log plot as follows:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/192/1*QeP8rM7GMRgoKL4wjQsoDA.png" /></figure><p>where <em>d</em> denotes the number of days since 1 January 2009. Estimating <em>b</em> yields 5.88 in my case. We can interpret this number as follows:</p><blockquote>As the time progresses 1%, the price increases 5.88% on average.</blockquote><p>Again, note the ambiguity of this interpretation. How can time move 1%? This time movement is fully dependent on the arbitrary starting date of 1 January 2009. Moreover, the further we move away from 2009, the more days fit into that 1% time interval. So we are not able to convert that 5.88% to average daily return or anything similar, giving us basically no understanding of the estimation results.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/952/1*WQUc-jvvumNMgCo1atjXGg.png" /></figure><h3>Stop using regressions</h3><p>My second argument is on the methodology of the analysis. He makes his prediction based on the linear approximation, which he obtains by means of a linear regression. I am strongly opposed to this particular method for this particular purpose. If you want to make price predictions, don’t use regressions, but just calculate some growth rate and extrapolate it into the future. Please consult my <a href="https://medium.com/amdax-asset-management/forecasting-the-bitcoin-price-using-holt-winters-exponential-smoothing-9a89f50d4ce1">previous article</a> on this particular topic.</p><p>Why does he still use a regression for his predictions? One reason is that it sounds sophisticated even though it’s very easy to implement. But another reason is so that he can use an R-squared. Generally speaking, an R-squared is a measure ranging from 0 to 1 that indicates how well the model explains the actual data (often referred to as the model fit). Or in terms of this case, how well the log-log plot is described by a linear relationship. But that is the general case, and this particular case happens to fall short of that…</p><p>Consider the following log-log plot together with the model’s R-squared from 2014 onwards. It is similar to the one presented in his article. The further we move along the time axis, the higher the R-squared. It is also very close to 1, its maximum value. His conclusion is that his model performs exceptionally well and even increasingly better over time. This is, as it appears, a false conclusion.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/995/1*igN9Gh-1Wz3dtjIWwGx0lg.png" /></figure><p>In statistics, there is a well-known case in which the R-squared of a linear regression approaches its maximum value of 1 as we add more and more observations, regardless of how good the model fit is. This happens if both variables (log-price and log-time) are 1) non-stationary and 2) not cointegrated.¹ Let’s go ahead and walk through both of these conditions.</p><h4>1) Non-stationarity</h4><p>A variable is non-stationary if its statistical properties (like mean and variance) are not constant over time. Non-stationarity holds true for any ever-increasing function, so the log-time is non-stationary by construction. For the log-price, things are a bit harder, so we turn to the Augmented Dickey-Fuller test. That is a statistical test that tells us whether a variable is stationary. I run the test using a multitude of sample periods. In particular, I set the starting date at 1 August 2010 and vary the ending date from 1 January 2019 until 1 August 2022 with weekly intervals. In the figure below, the relevant results are displayed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/950/1*wTdvgAmc-U-1w-OH6UaJhw.png" /></figure><p>The grey line is the 5% significance level, a commonly used critical level. If the test’s <em>p</em>-value is below this line, we reject non-stationarity and conclude that a variable is stationary. Note that the <em>p-</em>value falls below this line from May 2022 onwards but is usually well above the level. Hence, we are not able to firmly reject non-stationary and conclude that there are signs of non-stationarity in the log-price.</p><h4>2) Cointegration</h4><p>Cointegration is an abstract term that refers to two non-stationary variables moving around each other. It is intuitively explained by the example in this paper². To analyse the presence of cointegration, we turn to the Engle-Granger cointegration test. This test works in a similar way to the previous example. So we once again extract relevant <em>p</em>-values from the test over multiple sample periods, which are shown below. These paint a very clear picture. The <em>p</em>-values are well above the 5% significance level. So we do not reject the hypothesis that there is no cointegration between the two variables. In other words, there is no clear sign of cointegration.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/950/1*SnnEQqtxM9E1l18JvLOQsw.png" /></figure><p>Since both of the previously mentioned conditions hold, we are dealing with a so-called <em>spurious regression</em>. This spuriousness means that a linear regression has no value as it does not yield valid results. Hence, the increasing R-squared as reported in the article does not say anything about the fit of this model.</p><h3>Conclusion</h3><p>In this article, we have concluded that the power-law model is logically and statistically invalid. Logarithmically scaling time is irrational and has severe implications on the model as a whole. Furthermore, statistical theory renders the results from the linear regression useless.</p><p>Even though I disagree with his approach, I understand his goal. He seeks to adjust for the diminishing returns in the Bitcoin price series. After this adjustment, the price plot more or less resembles a linear relationship. But making something look nice and familiar doesn’t mean that it’s useful or legit. Especially not when the transformations are built upon ambiguous and arbitrary assumptions. We might as well just draw price predictions by hand while we’re at it.</p><h3>References</h3><p>[1] Phillips, P. C. (1986). Understanding spurious regressions in econometrics. <em>Journal of econometrics</em>, <em>33</em>(3), 311–340.</p><p>[2] Murray, M. P. (1994). A drunk and her dog: an illustration of cointegration and error correction. <em>The American Statistician</em>, <em>48</em>(1), 37–39.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1b40783657bf" width="1" height="1" alt=""><hr><p><a href="https://medium.com/amdax-asset-management/bitcoins-power-law-corridor-debunked-1b40783657bf">Bitcoin’s power-law corridor debunked</a> was originally published in <a href="https://medium.com/amdax-asset-management">Amdax Asset Management</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[All models are wrong, and this one isn’t even useful]]></title>
            <link>https://medium.com/amdax-asset-management/all-models-are-wrong-and-this-one-isnt-even-useful-a404fea38ca9?source=rss-865ada47c174------2</link>
            <guid isPermaLink="false">https://medium.com/p/a404fea38ca9</guid>
            <category><![CDATA[investing]]></category>
            <category><![CDATA[overfitting]]></category>
            <category><![CDATA[quantitative-finance]]></category>
            <category><![CDATA[quantitative-research]]></category>
            <category><![CDATA[bitcoin]]></category>
            <dc:creator><![CDATA[Tim Stolte]]></dc:creator>
            <pubDate>Tue, 16 Aug 2022 12:46:40 GMT</pubDate>
            <atom:updated>2022-08-16T12:46:40.718Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>“Quant investing is both art and science”</blockquote><p>This is how crypto influencer and <a href="https://medium.com/amdax-asset-management/stock-to-flow-a-philosophy-not-a-model-7d22c2eb7fd0">Stock-to-Flow</a> creator <a href="https://medium.com/u/bcb63a182704">PlanB</a> introduced his <a href="https://planbtc.com/20220807QuantInvesting101.pdf">new investing strategy</a> that he generously shared with his followers. Of course, the two-edged definition of quant investing allows him to present himself as an artist whose undeniable skill and wisdom will inspire all crypto investors left, right and center. There’s nothing new there. He has been doing this for years and he’s actually really good at it. It’s the other side that worries me though…</p><p>By stating that the quantitative approaches to investment decisions sometimes fall into the subjective category of art, he’s taking a hidden stance against all scientific principles of the subject. It effectively renders all logic and reason irrelevant once <em>the artist himself </em>says so. While I do think that quantitative methods can be creative, elegant and sometimes even mysteriously unexplainable, calling it art is another thing entirely and potentially dangerous. Data should be used intelligently and with care, because the past has proven time and again how easy it is to frame and deceive people with it. And this may very well be one of those times.</p><p>In this article, I will discuss what is wrong with PlanB’s article and strategy in a structured and detailed manner. Along the way, I’ll also zoom in on some quantitative and investing principles such that no prior knowledge on both topics is required. Let’s dive straight in!</p><h3>Trading rules</h3><p>In order to maximise profit as an investor, there are two things that you must do: buy when the price is low and sell when the price is high. Easier said than done of course, because we can only identify these perfect opportunities (when the price tops and bottoms) in hindsight. The concept of quant investing provides a possible solution to this problem by aiming to predict tops and bottoms using hidden patterns in data.</p><p>PlanB’s approach is very simple: use only the <a href="https://www.investopedia.com/terms/r/rsi.asp">Relative Strength Index</a> (RSI) in the model to pinpoint tops and bottoms in the Bitcoin price and thus find optimal buying and selling opportunities. The RSI is an index that takes a value between 0% and 100% to determine whether an asset is <em>overbought </em>(high RSI; indicating a top) or <em>oversold </em>(low RSI; indicating a bottom). So all we need to do is sell if the RSI is high and buy when the RSI is low. Simple right? Let’s take a look at the trading rules in the article:</p><ul><li>Sell if the RSI is below 65% now AND was above 90% in at least one of the last six months.</li><li>Buy if the RSI is more than 2% higher than the low AND was below 50% in at least one of the last six months.</li></ul><p>If your first reaction is that these trading rules seem a little complicated for such a simple metric, you’re right, but we’ll get to that later on. Instead, we’ll first focus on the implications of the signals. Consider the following graph, with the Bitcoin price in grey and the RSI in blue:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/986/1*r11OW6c4997VzzbJJxfFtA.png" /></figure><p>We immediately spot something odd. Buy and sell signals only appear when the RSI is relatively low. This is in sharp contrast with the interpretation of the metric, because we expected to sell for high RSI values which would signal a top. We even find that January 2012 is both a sell AND a buy signal. The strategy interprets it as a buy signal simply because we have <em>already</em> sold and cannot sell any more. In other words, we buy even though the strategy tells us to sell. In this situation, the signal depends on our current positioning rather than our expectation of where the price is heading. It basically flies in the face of the whole idea behind quant investing. But hey, art has no rules…</p><p>Some readers may counter these arguments along the lines of the good old saying: <em>“If it’s stupid and it works, it ain’t stupid”. </em>To those, I propose the following four-step method in true PlanB fashion:</p><h4>Data</h4><p>I have BTC monthly price data from <a href="https://glassnode.com/">Glassnode</a> and obtain year and month data from the Gregorian calendar. I use data from January 2011 until July 2022.</p><h4>Information</h4><p>From the BTC monthly price data, it looks like there are tops and bottoms that would perfectly correspond with long-term selling and buying opportunities, respectively.</p><h4>Knowledge</h4><p>It seems like those tops and bottoms occur more or less during the same months across multiple years.</p><h4>Wisdom</h4><p>We define two very simple trading rules that align with these opportunities:</p><ul><li>Sell at the end of December if the year ends with a 0, 3 or 7.</li><li>Buy at the end of January if the year ends with a 1, 5 or 9.</li></ul><p>Behold, the remarkable performance of this strategy, as depicted in the chart below. We start with 1 BTC in January 2011 and we end up with over 16.4 BTC in July 2022, beating the buy-and-hold strategy over 16.4 times. The drawdown of this strategy is -53% in BTC terms. It sharply outperforms PlanB’s strategy (9.7 BTC and -49% drawdown in BTC terms). Even though my strategy’s risk profile is a little more aggressive, the risk-adjusted returns remain higher by a wide margin.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/990/1*swsFU1M1U4jYB5n7pZKFiw.png" /></figure><p>I sincerely hope that everyone understands that these trading rules don’t make <em>any </em>sense and have <em>no interpretation</em>. And that even though it probably beats every conceivable long-term strategy, it should not be followed by anyone who is seeking to outperform a simple buy-and-hold strategy in the future. But interpretation is not the only thing missing from this type of investing approach, bringing us to the next topic.</p><h3>In-sample versus out-of-sample</h3><p>The article already mentions the term <em>out-of-sample performance</em>, but I am aware that it doesn’t immediately ring a bell with many. A quick introduction:</p><p>Consider two completely separate periods of price data, say period A and period B. We used period A to come up with some trading rules that yield amazing returns. As shown by the <em>long-term algorithmic exogenous quant investing strategy </em>example in the previous subsection, it is very much possible to create such rules for basically any data set with enough creativity. So showing that the rules work on period A is essentially meaningless. In order to successfully gauge the performance of a set of trading rules, they need to be applied to data points that are not the same as those that were used to create the rules. Hence, only period B can tell us how good our strategy actually is.</p><p>In this explanation, the data in period A is referred to as <em>in-sample</em> <em>data</em> and the data in period B is <em>out-of-sample data</em>. PlanB is well aware of the difference between the two, but unfortunately fails to mention the possible implications on his own analysis. So I will mention them here using a small example.</p><h4>Example</h4><p>Let’s take the liberty to remove all data from before July 2013. Then the BTC price vs RSI chart would have looked like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/986/1*tRUfw_JGcISRchVh7AaC2Q.png" /></figure><p>Now recall PlanB’s original sell rule that consisted of two parts: sell if i) the RSI is below 65% now AND ii) the RSI was above 90% in at least one of the last six months. Based on the new in-sample data set, this sell rule would make no sense, because we can simplify the rule and gain a lot of performance. We simply remove part i) of the rule and adjust part ii), such that our new sell rule is as follows: sell if the RSI is above 95%.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/980/1*256SYP1wcXpmnonQRzWinw.png" /></figure><p>We find that this new strategy (in green) yields returns that are almost three times as high as the returns of the old strategy (in blue): 14.4 BTC vs 5.6 BTC. Moreover, the risk (as measured by drawdown) is significantly less: -30% vs -49%. The red line corresponds with the strategy that sells if the RSI is above 90%. This one yields reasonable returns, but is nowhere near optimal. Hence, for this particular data set, the RSI &gt;95% sell strategy is definitely the way to go.</p><p>But here is the catch: how does this strategy perform if we add the out-of-sample data from January 2012* to July 2013 back in the mix. See for yourself in the graph below. Evidently, the performance before July 2013 is so bad that our new strategy just barely outperforms buy-and-hold (which equals 1 on the right axis) once we consider the full picture. This is the power of out-of-sample performance and the reason that it should always be included in strategy testing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/974/1*jeW1-leta9CGXDfo89RxTQ.png" /></figure><p>*<em>I ignore data from 2011 on purpose because of the extremely high and inaccurate RSI calculations in that year, causing our new strategy to immediately sell and yield practically zero returns. Since this would be an unfair way to prove my point, I start in January 2012.</em></p><p>In this subsection, I showed that <em>only </em>the use of out-of-sample analysis can judge whether a trading strategy performs well in general. The above example shows a strategy that performs exceptionally well at first, but totally fails once we add one and a half years of out-of-sample data. And no sane investor would bet his money on it, no matter how good the in-sample performance was. In conclusion, the fact that PlanB’s strategy outperforms buy-and-hold 10 times on an in-sample basis does not say <em>anything </em>about how well it would actually perform overall.</p><h3>Overfitting</h3><p>The third and final point of criticism I will put forward is very closely related to the previous one. For a very good example on what overfitting is and how dangerous it can be, please consider <a href="https://ml.berkeley.edu/blog/posts/crash-course/part-4/">this article</a> on how it caused the Fukushima disaster. PlanB himself describes the general concept fairly well:</p><blockquote>Overfitting happens when an algorithm memorizes a dataset (including the noise) instead of generalizing the underlying signal. Overfitting gives good model performance when making a model but poor model performance on new out-of-sample data. Preventing overfitting is an art in TA, statistics and AI.</blockquote><p>Ironically enough, while he is familiar with the definition and apparently even considers himself <em>an artist in preventing overfitting, </em>he still totally fails to do so. His desperate attempt to prevent it is to use only one metric in order to keep the model as simple as possible. Sometimes that approach is sufficient, but overfitting also happens if you make your model too dependent on the data of that one particular metric. Remember when I said that his trading rules seemed a little complicated for a simple metric such as the RSI? That’s exactly where he started overfitting. Here’s why:</p><h4>Sell signal</h4><blockquote>Sell if the RSI is below 65% now AND was above 90% in at least one of the last six months.</blockquote><p>In the previous subsection on out-of-sample performance, I showed that the “<em>below 65% part</em>” of this sell rule is solely based on the <em>double top</em> in 2013. Specifically because, if we ignore the data in early 2013, we find that there is absolutely no need for that part whatsoever. If that is not memorising the data set, then I don’t know what is.</p><p>Ignoring the “<em>below 65% part</em>”, I would also argue that the 90% in the second part of this rule is chosen <em>only</em> because the single top in March 2021 just barely missed out on exceeding 95% just like all other tops. Let’s say we didn’t have data from March 2021 onwards and created trading rules. We would have definitely chosen the 95% RSI barrier, because its performance would have been two times better. But nowadays we know that that would have caused us to miss the selling opportunity in March 2021. So we choose 90% instead of 95% based on one observation only. What’s that called? Oh yeah, overfitting.</p><h4>Buy signal</h4><blockquote>Buy if the RSI is more than 2% higher than the low AND was below 50% in at least one of the last six months.</blockquote><p>Now consider the following statement made in the article:</p><blockquote>Traditionally, RSI above 70 indicates an overbought situation and RSI below 30 indicates an oversold condition. However, the BTC range is different because BTC RSI can go as high as 90–100 and has never been lower than 40.</blockquote><p>We now know that the RSI of traditional assets can go below 30%. And we are already in buying territory if the Bitcoin RSI is at 50%? It seems foolish to think that an RSI around 30% will not happen in the coming years, just because in the one case of Bitcoin it didn’t happen yet. We have also never seen the Bitcoin price exceed $70k, but many of us still think the chance of this happening eventually is quite high, right? Once again, we’re memorising the data instead of thinking about what it actually means.</p><h3>Conclusion</h3><p>In this article, I reviewed PlanB’s strategy that uses the Relative Strength Index to construct trading rules that would have yielded outstanding returns in hindsight. And based on the interpretation of the trading rules, lack of out-of-sample analysis, and the obvious presence of overfitting, I can undoubtedly state that this strategy should not be considered a useful tool to possibly outperform a simple buy-and-hold strategy in the future. Sure, there is a reasonable chance that it <em>can </em>outperform, but not by virtue of the model’s logic.</p><p>Luckily, PlanB appears to be prepared for the case in which his strategy fails or is debunked. He introduces a (fairly substantial) list of disclaimers corresponding with his four steps:</p><p><strong>Data: the data could contain errors. </strong>Sure, but since we only use price data we should be fine. It is probably the most accurate data in crypto.</p><p><strong>Information: the calculation of RSI could be wrong. </strong>Yes, which is why quants should always perform their own RSI calculation.</p><p><strong>Knowledge: the correlation between RSI and BTC could be spurious. </strong>Impossible, the RSI is based on price data and nothing else. By construction, there is correlation between the two. Fact.</p><p><strong>Wisdom: overfitting of the trading rule, the backtest could be wrong, there could be black swans and past performance is no guarantee of future results. </strong>It is the responsibility of the quant to eliminate the risks of overfitting and incorrect backtesting. This basically says “I may not have done my job correctly”. The black swans part is true, but only for <a href="https://www.forbes.com/sites/johndrake/2021/11/11/was-covid-19-a-black-swan-event/?sh=dc353f4bd36a">actual black swans</a>. Lastly, the past performance part aligns with my points, but should have been included in the analysis instead of the disclaimers.</p><h3>Appendix</h3><p>I used the following code to calculate the Relative Strength Index:</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/bb8d96510b9e38f197f75746c930b7a4/href">https://medium.com/media/bb8d96510b9e38f197f75746c930b7a4/href</a></iframe><p>If you want to review my backtesting method, I’ve created a Python class specifically for the purpose of this article.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/a02515afcfe1b56ad86a66ec8bb007e3/href">https://medium.com/media/a02515afcfe1b56ad86a66ec8bb007e3/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a404fea38ca9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/amdax-asset-management/all-models-are-wrong-and-this-one-isnt-even-useful-a404fea38ca9">All models are wrong, and this one isn’t even useful</a> was originally published in <a href="https://medium.com/amdax-asset-management">Amdax Asset Management</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Correlation Controversy]]></title>
            <link>https://medium.com/amdax-asset-management/the-correlation-controversy-395a3cdf27?source=rss-865ada47c174------2</link>
            <guid isPermaLink="false">https://medium.com/p/395a3cdf27</guid>
            <category><![CDATA[crypto]]></category>
            <category><![CDATA[econometrics]]></category>
            <category><![CDATA[portfolio-management]]></category>
            <category><![CDATA[correlation]]></category>
            <category><![CDATA[quantitative-research]]></category>
            <dc:creator><![CDATA[Tim Stolte]]></dc:creator>
            <pubDate>Mon, 08 Aug 2022 20:06:15 GMT</pubDate>
            <atom:updated>2022-08-08T20:06:15.297Z</atom:updated>
            <content:encoded><![CDATA[<h4>Part III: Crypto Correlations</h4><p>This article series aims to answer the question of whether crypto moves independently from traditional assets in such a way that it can be considered its own asset class. If that is the case, investors can use crypto to diversify their multi-asset portfolios and limit their exposure to extreme market movements, as discussed in <a href="https://medium.com/amdax-asset-management/the-correlation-controversy-6986e35a34be">Part I</a>. Afterwards, in <a href="https://medium.com/amdax-asset-management/the-correlation-controversy-e6ebcbd742f1">Part II</a>, we encountered multiple difficulties than come with measuring correlation. Being aware of these complications, we may better understand to what extent crypto and other asset classes are intertwined. We wrap up our search for this answer in the third and final part of this series.</p><h3>Data</h3><p>Let’s start off by introducing the data that we will be using for our analysis. We use the Bitwise 10 Crypto Index Fund as a proxy for the crypto market. The fund is comprised of the 10 largest cryptocurrencies, weighted by market capitalization and rebalanced monthly, making it a valid choice. Besides crypto, we will be investigating four other asset classes and their corresponding proxies:</p><ul><li>Equities; The MSCI World Index captures the performance of global large and mid-cap equities across 23 developed market countries.</li><li>Fixed Income; The SPDR Bloomberg Global Aggregate Bond Index ETF consists of a broad range of government and corporate bonds and securities.</li><li>Commodities; The iShares S&amp;P GSCI Commodity-Indexed Trust tracks the performance of various commodities, such as energy, industrial and precious metals markets.</li><li>Real Estate; The Vanguard Real Estate ETF measures performance of real estate-related investments.</li></ul><p>The scaled time series, running from 4 January 2021 until 29 July 2022, are displayed below. Right off the bat, we observe a major difference in volatility. Crypto reaches higher highs and lower lows than any of the four other asset classes. As expected, the bond market shows the smallest movements, making it especially suitable for decreasing volatility in a portfolio. One striking observation is that equities and real estate appear to move in tandem over the full sample period. And particularly at the end of the sample period, crypto seems to follow a similar pattern.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/933/1*fql9jWP-xLi46kmMqr5Aqg.png" /></figure><h3>Correlation analysis</h3><p>From the previous article, we know that judging correlation by eye can lead to false conclusions, so we must dive deeper into the data. We ignore the raw values (for reasons given in the previous article) and use day-to-day changes to obtain sample correlation coefficients. These are shown in the correlation matrix below. We find that there is a high degree of independency across most asset classes. Apart from two exceptions, all correlations are below 0.2, which can be considered very low. This confirms that we may be dealing with separate asset classes across our sample set. One might argue that the significant positive correlation of 0.74 between real estate and equities refutes the statement that real estate should be considered its own asset class. But the same may not be said about crypto, showing only a 0.22 correlation coefficient with equities.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iLGf4MeYGMsXafXPv1-vXw.png" /><figcaption>Correlation matrix of day-to-day changes for our five asset classes over the sample period 4 January 2021–29 July 2022</figcaption></figure><p>We know that day-to-day changes may not capture the full picture. We therefore also consider week-on-week changes. The results are more or less the same, except that fixed income correlations with equities and real estate have ramped up considerably. This suggests that bonds may not be the safe haven they were in past decades. Again, we observe that the crypto correlations are very low.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xeKXoEbhkFpsmm3a3y6xLw.png" /><figcaption>Correlation matrix of week-on-week changes for our five asset classes over the sample period 4 January 2021–29 July 2022</figcaption></figure><p>The correlation matrices make the critical assumption that correlation is constant across our sample period. We have seen in <a href="https://medium.com/amdax-asset-management/the-correlation-controversy-e6ebcbd742f1">Part II</a> that relaxing this restriction may provide us with more insights. From now on, we will only focus on correlations between crypto and the four other series. The graph below displays the 90-day rolling window correlations of crypto and each of the four other asset classes. These correlations are highly time-varying, showing negative values during mid-2021. Afterwards, crypto becomes increasingly correlated with both equities and real estate, which we know are heavily interconnected. It is important to note that at the end of our sample period, correlation with equities, fixed income and real estate is well above 80 and even exceeds 90 in the last couple of days. This suggests that crypto has considerably lost its diversifying power over the course of last year. Of course, there is no way of knowing whether this will persist, but it is good to be aware that exceptionally high correlation is not a rarity.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/944/1*ZLTOGHh625jjb6MIF4lyuA.png" /></figure><h3>Interpreting the results</h3><p>Can we think of an explanation as to why the correlation between crypto and equities reached such high levels in the past few months? After looking at the data, it seems to be that the downturn in equity prices may be one of the possible causes for this to happen. The graph below shows where this suspicion comes from. It is very clear that once equity prices started to fall in January 2022, the correlation increased from 0 to over 0.9. A similar feature can be found during the short downturn in September and October 2021.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/977/1*qdR4yRRuOYPOD5Btp8ES3Q.png" /></figure><p>To find out whether the relationship between crypto and equities actually becomes stronger in an economic decline, let’s consider the scatter plot below. It distinguishes between dates on which equity returns are positive (blue) and dates on which they are negative (red). A linear regression yields R-squared values, which give an indication of how strong the relationship between crypto and equity returns are. Note that the relationship during periods of negative equity returns is significantly higher than during periods of positive equity returns. Nevertheless, since the R-squared values are low, we can conclude that this distinction only explains a small part of the positive correlation between the two asset classes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/628/1*EJDP5p-z98XKHx261K6JyQ.png" /></figure><p>Another approach of explaining the correlation would be to consider extreme market conditions. We can, for instance, define such conditions as dates on which the equity returns are highest or lowest. The scatter plot below shows what that looks like, with the 10% most extreme equity return dates in red. Looking at the R-squared of these red observations, we immediately spot that the relationship is better explained than in the previous case. Crypto returns are generally negative during exceptionally low equity returns and positive during exceptionally high equity returns. The R-squared of 0.33 more or less confirms this observation. We thus draw the conclusion that the high correlation between crypto and equities at the end of our sample period may, to a certain extent, be explained by extreme market conditions and volatile price movements.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/628/1*ZdMra_NIVPz5hFMdGyK43g.png" /></figure><h3>Final remarks</h3><p>We now know what is going on between crypto and equities. Our analysis on the data since January 2021 tells us that the correlation between the two has been low or even sharply negative. But in the past few months, it settled above 90% in what looks to be a steady and decisive manner. Does this then revoke crypto’s right to be considered a diversifying asset? Not necessarily, because two other asset classes, fixed income and real estate, are also heavily correlated with equities during this time. Furthermore, we don’t know whether this correlation will continue indefinitely. We have reasons to believe that this strong interconnection between crypto and equities is caused by the economic downturn that held the markets in its grip in the first two quarters of 2022. It thus appears that the state of the global economy affects the behaviour of crypto assets relative to other assets.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=395a3cdf27" width="1" height="1" alt=""><hr><p><a href="https://medium.com/amdax-asset-management/the-correlation-controversy-395a3cdf27">The Correlation Controversy</a> was originally published in <a href="https://medium.com/amdax-asset-management">Amdax Asset Management</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Correlation Controversy]]></title>
            <link>https://medium.com/amdax-asset-management/the-correlation-controversy-e6ebcbd742f1?source=rss-865ada47c174------2</link>
            <guid isPermaLink="false">https://medium.com/p/e6ebcbd742f1</guid>
            <category><![CDATA[quantitative-research]]></category>
            <category><![CDATA[statistics]]></category>
            <category><![CDATA[crypto]]></category>
            <category><![CDATA[correlation]]></category>
            <dc:creator><![CDATA[Tim Stolte]]></dc:creator>
            <pubDate>Fri, 05 Aug 2022 12:39:47 GMT</pubDate>
            <atom:updated>2022-08-30T12:03:50.258Z</atom:updated>
            <content:encoded><![CDATA[<h4>Part II: The Problem</h4><p>In <a href="https://medium.com/amdax-asset-management/the-correlation-controversy-6986e35a34be">Part I</a> of this article series, I discussed the relevance of defining a crypto asset class and why the correlation of price dynamics plays a relevant part in this context. This article will dive into the complexities of correlation. I will explain why it is so hard to come up with a single measure that tells us to what extent two time series interact with each other.</p><h3>An example</h3><p>To best illustrate the problem, consider the two self-generated (synthetic) daily time series shown in the figure below. At first glance, they look more or less perfectly correlated. Calculating a sample correlation coefficient using the raw (untransformed) series gives 0.97. This fully supports our expectation, so it should be correct, right? Well, no. This approach misses out on one crucial aspect which can lead to not only an unnuanced conclusion, but even a completely incorrect one. I will explain the caveat using two different arguments: empirical and theoretical.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/928/1*3olm4Aim6CmJHkToNRlRoQ.png" /></figure><h4>Empirical argument</h4><p>For the empirical line of thinking, let’s try to work out what correlation actually means. Start by taking all the days on which y1 increases. If y1 and y2 are positively correlated, we of course expect y2 to also increase on most of those days. Negative correlation would imply that y2 generally moves in the opposite direction of y1. If the series are independent, y2 increases on around half of the days and it decreases on the other half. A similar logic obviously also applies to days on which y1 decreases.</p><p>Let’s apply this to our example. y1 increases on 295 out of 577 days. And out of these 295 days, y2 only increases on 79 days. In other words, y2 only<strong> </strong>increases on 27% of all days on which y1 increases. Similarly, y2 also only decreases on 26% of all days on which y1 decreases. This means that the two time series move in opposite directions almost 75% of the time. Anyone would thus conclude that they are more negatively than positively correlated.</p><h4>Theoretical argument</h4><p>The theoretical explanation boils down to one of the underlying assumptions of the formula for calculating a sample correlation coefficient. In order for the formula to be reliable, both time series must have a time-invariant sample mean. In other words, the mean of the time series should be the same for all subperiods within the full sample period. This is clearly not the case for our synthetic series. Their sample mean is somewhere between 0 and 5 in the first four months and ends up well above 20 at the end of the sample period.</p><p>To put it even simpler: the sample correlation coefficient effectively calculates how often both series are on the same side of their sample mean at the same time. Now, y1 and y2 share the same sample mean at around 14.5. We can easily see that both series are both below 14.5 in the first two months and above 14.5 in the last three months. Evidently, applying this measure of correlation seems highly unnuanced, as any two series which happen to trend in the same direction would be almost perfectly positively correlated.</p><h3>Searching for a solution</h3><p>Both the empirical and theoretical reasoning point in the same direction. In order to identify true correlation, we should consider day-to-day changes rather than the actual raw values. Using first differences in the sample correlation coefficient yields a value of -0.69, confirming that y1 and y2 are negatively correlated.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/928/1*HArV60PiI7ZeYvIS4PfGCw.png" /></figure><p>Regrettably, this seemingly easy fix comes with its own problem. Because even though y1 and y2 generally trend upwards in the long-term, they both experience a downtrend at the same points in time. This has to account for something in our search for correlation, right? Taking first differences distorts our view such that we are no longer able to recognise these features on the medium-term. To fix this “short-sightedness”, let’s take month-on-month differences and see what happens with the correlation coefficient. It now equals 0.24. Hence, on a monthly basis, we find that y1 and y2 move much more independently from each other.</p><p>So what is it? Are y1 and y2 negatively correlated, independent, or were they positively correlated all along? The answer probably lies somewhere in the middle. It can even vary over time. This is the reason that a lot of analysts turn to rolling correlations. Such an approach would look something like this in our case:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/938/1*NFafXoEDyPEEmeMeu3fCRQ.png" /></figure><p>Clearly, rolling correlations can vary significantly over time, with the 30-day correlations ranging somewhere between -0.85 and -0.45. This makes it practically impossible to pinpoint a number that captures the co-movement of two time series over a certain time period. Moreover, this approach introduces additional ambiguity to the measure, because we can opt for rolling window sizes of 30 or 90, but any other number might also be a valid choice. And of course, no one knows which size is correct and any choice has its own implications.</p><h3>Conclusion</h3><p>To side with the critical reader, I definitely constructed these time series such that they would yield the seemingly contradictory result I needed to prove my point. But the fact that I am able to do so is enough to demonstrate how complex the principle of correlation can be in extreme cases. After all, there is no reason to believe that the unpredictable world we live in cannot produce something as paradoxical as this Python script.</p><p>In this article, we have touched upon the most basic aspects of correlation. Transforming raw series into differences validates the use of the sample correlation coefficient formula. But for a good correlation analysis, we should consider multiple timeframes and maybe even time-varying properties. But which complexities apply to real world problems? In particular, how do the findings in this article help us to identify the dependence or independence between crypto and other asset classes? We’ll find out in the <a href="https://medium.com/amdax-asset-management/the-correlation-controversy-395a3cdf27">next part</a>.</p><h3>Appendix</h3><p>In this simulation study, I created two synthetic time series according to the following equations:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/1*hoyB9OR9x7cfHLR79o3A2w.png" /></figure><p>The code that can be used to reproduce my findings can be found below.</p><pre>from datetime import datetime<br>import pandas as pd<br>import numpy as np</pre><pre>dates = pd.date_range(datetime(2022, 1, 1), datetime(2022, 7, 31))<br>n = len(dates)<br>random_seed = 3<br>rng = np.random.default_rng(random_seed)<br>mu = np.array([0, 0])<br>sigma = np.array([[1, -1],<br>                  [-1, 2]])<br>epsilon = rng.multivariate_normal(mu, sigma, size=n)<br>t = 0.05*np.arange(0, n)<br>y1 = pd.Series(t + 2*np.sin(t) + epsilon[:, 0], index=dates)<br>y2 = pd.Series(t + 2*np.sin(t) + epsilon[:, 1], index=dates)</pre><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6ebcbd742f1" width="1" height="1" alt=""><hr><p><a href="https://medium.com/amdax-asset-management/the-correlation-controversy-e6ebcbd742f1">The Correlation Controversy</a> was originally published in <a href="https://medium.com/amdax-asset-management">Amdax Asset Management</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Correlation Controversy]]></title>
            <link>https://medium.com/amdax-asset-management/the-correlation-controversy-6986e35a34be?source=rss-865ada47c174------2</link>
            <guid isPermaLink="false">https://medium.com/p/6986e35a34be</guid>
            <category><![CDATA[quantitative-research]]></category>
            <category><![CDATA[portfolio-management]]></category>
            <category><![CDATA[crypto]]></category>
            <category><![CDATA[diversification]]></category>
            <category><![CDATA[correlation]]></category>
            <dc:creator><![CDATA[Tim Stolte]]></dc:creator>
            <pubDate>Wed, 03 Aug 2022 09:21:21 GMT</pubDate>
            <atom:updated>2022-08-30T12:03:07.881Z</atom:updated>
            <content:encoded><![CDATA[<h4>Part I: A Crypto Asset Class</h4><p>Over the past few years, a multitude of perspectives have incentivized investors to expose their portfolios to the cryptocurrency domain. Probably the most straightforward argument considers the underlying blockchain technology as one of the most revolutionary developments in modern financial history. It essentially portrays cryptocurrencies as technology assets which are prone to above-average growth, much like most NASDAQ assets in traditional finance. Then there is the investor type that views the crypto market as the ideal opportunity to place their asymmetric bets. And while the huge upside potential is undeniably present, this argument is often disguised as the first one in hilarious attempts to deny greediness. But there is one more major reason that may dwarf the others in just a few years time.</p><blockquote>An asset class is a set of assets that bear some fundamental economic similarities to each other, and that have characteristics that make them distinct from other assets that are not part of that class — Robert J. Greer, 1997</blockquote><p>At the heart of investing lies the principle of asset classes. Some well-known examples are equities (e.g., stocks), fixed income (e.g., bonds) and commodities (e.g., precious metals). They partly determine the characteristics of an investment portfolio and reflect its underlying strategy. Think, for instance, of the famous 60/40 rule, which states that a good portfolio is allocated to both stocks and bonds for 60% and 40% respectively. Rules like this ensure that a portfolio is well-diversified, ideally limiting exposure to downward price movements.</p><p>The question now arises whether crypto should be considered a new asset class. Or in other words, whether the shared characteristics of the assets across the crypto space make them sufficiently different from other assets like stocks and bonds. In the context of governance this is definitely the case. This role is fulfilled by many different entities, for instance, blockchain and/or application developers, miners, validators, and many more. Obviously, there are no similarities here with any other asset class. Asset supply is also a crucial aspect in which significant differences can be found. If we, for example, take a look at the foreign exchange asset class, central banks hold the power to adjust the money supply. This enables them to increase or decrease the value of the currency that is held by the public. We find similar features for other asset classes, where companies can issue or buy back stocks and bonds, thus controlling their market value and affecting portfolios of investors. In contrast, crypto assets are subject to a predetermined supply schedule that is defined by the underlying protocol. And while some protocols may change over time, any adjustments are open-source and transparent. There are many more reasons to support the statement that crypto assets form their own asset class based on fundamental differences, for which I refer to the book <em>Cryptoassets </em>by Burniske and Tatar.¹</p><h3>Where correlation comes in</h3><p>So far, we have argued from a theoretical perspective that crypto is its own asset class. But is that perspective enough? For some individuals it might be, but for full acknowledgement, we need the practitioners on our side. They are after all the ones who are building portfolios, seeking the highest returns for the lowest risk.</p><p>Once again consider the 60/40 rule from earlier. In order for this rule to actually work (i.e., diversify the investments), it is important for the considered asset classes to actually behave differently. If your 60% allocation in stocks drops, at the very least you don’t want your 40% bond allocation to go down with it.</p><p>How do you gauge whether an asset mix adheres to this principle? Would combining crypto assets and other asset classes result in a well-diversified portfolio? The fundamental basics we discussed before have some impact on price movements, but they do not guarantee that the dynamics are fully independent. We also need to take into account that any hidden (or less obvious) drivers can impact two totally separate asset classes in the same manner. In the end, we want to find some measure that accurately captures the actual relationship between multiple price series. And contrary to popular belief, it is quite hard to do that.</p><p>Read Part II <a href="https://medium.com/amdax-asset-management/the-correlation-controversy-e6ebcbd742f1">here</a>.</p><h3>References</h3><p>[1] Burniske, C. &amp; Tatar, J. (2018). <em>Cryptoassets: The innovative investor’s guide to bitcoin and beyond</em>. New York: McGraw-Hill Education.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6986e35a34be" width="1" height="1" alt=""><hr><p><a href="https://medium.com/amdax-asset-management/the-correlation-controversy-6986e35a34be">The Correlation Controversy</a> was originally published in <a href="https://medium.com/amdax-asset-management">Amdax Asset Management</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Forecasting the Bitcoin price using Holt-Winters Exponential Smoothing]]></title>
            <link>https://medium.com/amdax-asset-management/forecasting-the-bitcoin-price-using-holt-winters-exponential-smoothing-9a89f50d4ce1?source=rss-865ada47c174------2</link>
            <guid isPermaLink="false">https://medium.com/p/9a89f50d4ce1</guid>
            <category><![CDATA[econometrics]]></category>
            <category><![CDATA[quantitative-research]]></category>
            <category><![CDATA[bitcoin]]></category>
            <category><![CDATA[exponential-smoothing]]></category>
            <category><![CDATA[holt-winters]]></category>
            <dc:creator><![CDATA[Tim Stolte]]></dc:creator>
            <pubDate>Wed, 06 Jul 2022 18:44:15 GMT</pubDate>
            <atom:updated>2022-07-06T18:44:15.946Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>Looking at the Bitcoin price chart belongs to any crypto fanatic’s daily routine. It’s often nothing more than just a glance, like checking an email inbox or refreshing a Twitter feed. Occasionally, however, one might engage in the daunting act of speculation. This would generally involve drawing an imaginary line on top of a historical price chart, aiming to capture some growth rate and extending that into the future. Many crypto analysts have made a habit of overcomplicating this exact approach. They also like to refer to it with a fancier name: <em>linear time series regression</em>.</p><p>The problem is not necessarily with the name. It is indeed some sort of linear regression, as it relates a variable of interest (the Bitcoin price) to some other variable (time). Regrettably though, such regressions are often used with very little regard for any implications¹ that the lack of statistical properties may have on the validity and interpretation of the estimation results. Using a regression to obtain a rough prediction of the Bitcoin price is a bit like using a shotgun to kill a fly. It gets the job done, but it also completely destroys the wall upon which the fly was minding its own business.</p><p>Used correctly, regressions can provide an analyst with indicators that gauge the accuracy of the estimates. But even if that’s the case, most people are not looking for confidence intervals with a width equal to that of the sun. I therefore firmly believe that regressions should not be used for long-term price predictions.</p><p>What should we use then? Well, a few weeks ago, I was approached by <a href="https://medium.com/u/6ca598f82fcf">Will Clemente</a> to peer review his team’s recently published <a href="https://newsletter.blockwareintelligence.com/p/bitcoin-user-adoption-report?s=r">Bitcoin User Adoption Report</a>². The piece introduced a way to predict the evolution of the Bitcoin adoption using the historical adoption curves of major technologies. Their hypothesis is that a weighted average of these curves provides a reasonable guess from the point of 10% global adoption onwards. Before that 10% adoption is reached, the cumulative sum of net growth of entities is used to proxy for the actual number of people that are holding Bitcoin at any point in time. Future values of this metric are predicted using a very simple extrapolation. No roughly drawn lines on a chart, no exogenous factors in a linear regression, just basic exponential extrapolation. That seemed like a great idea…</p><h3>Holt-Winters Exponential Smoothing</h3><p>Exponential smoothing is a rather well-known and elegant solution for generalising data into a single formula, ready to use for forecasting. Besides its flexibility, I am exploiting the advantage that no statistics (i.e., stochastic components) are involved in this approach. Hence, the method doesn’t require us to make any assumptions apart from the specification itself.</p><p>Back in 1957 and 1960, C. E. Holt³ and P. R. Winters⁴ published two individual papers that together introduced a novel way of applying exponential smoothing. This newly found method broke down a time series into three different components: level, slope and seasonality. I will briefly discuss each component individually, building towards the final set of equations along the way.</p><p><strong>Level: </strong>At any point <em>t</em> in time, the level of the time series equals the expected value of all future points<em>. </em>So our expectation of tomorrow’s value will be equal to the level, but also our expectation of the value one week or even one year from now. In general:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/80/1*nVLEQnv0Rp0kfboc4R-L6Q.png" /><figcaption>Main equation v1.0</figcaption></figure><p>where <em>h</em> denotes the number of points we are looking ahead in time. The <em>|t</em> means that <em>t</em> is our current timestep and all information up until that point is available to us. We also define an equation that describes how we obtain the level based on current and previous data:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/171/1*TqXWNbWSfCQ55XF1TPJUIw.png" /><figcaption>Level equation v1.0</figcaption></figure><p>Hence, the current level (generalised expectation of future values) is a combination of the current value and the previous level.</p><p><strong>Slope: </strong>The slope is a measure of how much the time series value changes from one timestep to another. We include this in our main equation using a growth factor <em>g</em> and the number of timesteps into the future <em>h</em>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/127/1*TkPDw1CZVpfL9U77Hh51FQ.png" /><figcaption>Main equation v2.0</figcaption></figure><p>The slope equation again follows from a linear combination, but this time of the level difference and the previous growth factor:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/236/1*XbrtQMVeQLQTaqgNDjicFw.png" /><figcaption>Slope equation v2.0</figcaption></figure><p>After introducing the slope component, we are required to adjust the level equation to account for the level difference:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/237/1*TEH0J17Y7YWkRdj77daFZA.png" /><figcaption>Level equation v2.0</figcaption></figure><p><strong>Seasonality: </strong>Seasonality adds the last level of complexity to our system of equations. It basically searches for a historical point in time that resembles the point that we aim to predict. For example, if we are forecasting a future value in January in a monthly time series, seasonality would allow us to incorporate specific information of last January. For weekly time series, we can think of looking back at the previous Monday if we are predicting a future Monday value.</p><p>In our case, we are looking for a 4-year seasonality. This would incorporate information on the Bitcoin halving cycle that is prominently present in the Bitcoin price series. In particular, I use a additive seasonality component, as I find that the seasonal variations of the last two cycles were more or less similar in magnitude. I will not go into detail on the differences between additive and multiplicative versions of the Holt-Winters method but refer to <a href="https://medium.com/analytics-vidhya/a-thorough-introduction-to-holt-winters-forecasting-c21810b8c0e6">this article</a>⁵ for a clear and concise description. Our final set of equations (along with seasonal adjustments in the main and level equations) looks like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/339/1*2hHpjc_3DH345iO-7J7Zww.png" /></figure><p>In this system, <em>m</em> is denoted as the number of timesteps that make up one season. Note that the level, slope and seasonality component are all included in the main equation and that each component is described by a linear combination with parameters α, β and γ, which are parameters to be estimated.</p><h3>Forecasting the Bitcoin price</h3><p>In my analysis, I use monthly closing price for Bitcoin from December 2013 until June 2022. I have multiple reasons for this choice of data. Firstly, monthly data is a better choice for a long-term price prediction than weekly or daily data. Secondly, I feel that the dynamics of price or on-chain data before 2014 can be unrepresentative for the remainder of the data due to the lack of adoption and highly inconsistent consumer behaviour at the time. Lastly and probably most importantly, I like how the chart looks. My sample period contains two full halving cycles, and the price dynamics align with the assumptions in my Holt-Winters specification.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/947/1*l8jfGdPL7WZtcZOukO5p-Q.png" /></figure><p>After applying the Holt-Winters method as described in the previous section, using the natural logarithm of the price as the variable of interest and <em>m</em>=48, we find the following smoothed time series:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/947/1*KrljozLaGYqm26_jVrEfMQ.png" /></figure><p>This is often the point at which model advocates would have come up with an R²<em> </em>or some other performance measure to give an idea of how accurate their approximation is. Truth be told, I have no idea how good this smoothing algorithm is at this point. And any measure that one may come up with has little to no interpretation. The main reason for this is that we have used the information of the full sample period to fit our parameters to the data. In that case, we might as well have gone with an approach that sets our prediction equal to the actual value at each point in time, report an R²<em> </em>of 1 and call it a day.</p><p>Still, smoothing looks pretty satisfying. Let’s see what happens when we leave some data points out for out-of-sample predictions. We start by applying Holt-Winters to the data sample up until December 2021, use it to make a one-month-ahead forecast and then use that forecast in the next iteration to make another forecast. We repeat this process until we have forecasts for the first six months of 2022:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/947/1*A1U7mm0txRS8itU1cQduEA.png" /></figure><p>We find that both the predicted values are slightly different from the smoothed values that we have seen before. This is a direct consequence of ignoring the last six months in estimating our parameters. The predictions also appear to drop off quite aggressively, even though such a downward trend was not yet confirmed as of December 2021 (the monthly closing price was still around $46,000). Nevertheless, it turns out that this downturn was quickly justified as the price also experienced a sharp decrease.</p><p>Unfortunately, estimation in our current setup requires at least two full seasons, so we cannot enlarge our hold-out sample with the current sample period. Whether the Holt-Winters method statistically outperforms other approaches on an out-of-sample basis remains a question for future research.</p><h3>A look into the future</h3><p>What happens if we forecast values on a longer term, something like one cycle into the actual future? The figure below shows what that looks like. We observe a pattern very similar to the one we have seen in previous halving cycles, indicating that the seasonality component works its magic.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/947/1*oxIK_52IpxhVnLwLWojdqw.png" /></figure><p>I must stress that these predictions are nothing more than educated guesses, and any confidence interval drawn in this chart would be statistically invalid. Of course, the further away the forecast, the more sceptical one should be about its accuracy. The limited number of data points also negatively impacts the robustness of the model. For instance, the current predictions find the next top at around $248,000 in December 2025. But if next month’s closing hits anywhere above $25,000 (we just closed June 2022 at $19,800), Bitcoin is bound break the $300,000 mark before 2026 according to Holt and Winters.</p><h3>References</h3><p>[1] Stolte, T. (2022). Stock-to-Flow: a philosophy, not a model. <a href="https://medium.com/amdax-asset-management/stock-to-flow-a-philosophy-not-a-model-7d22c2eb7fd0">https://medium.com/amdax-asset-management/stock-to-flow-a-philosophy-not-a-model-7d22c2eb7fd0</a></p><p>[2] Blockware Intelligence Team (2022). Bitcoin User Adoption Report. <a href="https://newsletter.blockwareintelligence.com/p/bitcoin-user-adoption-report?s=r">https://newsletter.blockwareintelligence.com/p/bitcoin-user-adoption-report?s=r</a></p><p>[3] Holt, C. E. (1957). Forecasting seasonals and trends by exponentially weighted averages (ONR Memorandum №52). <em>Carnegie Institute of Technology, Pittsburgh USA</em>, <em>10</em>.</p><p>[4] Winters, P. R. (1960). Forecasting sales by exponentially weighted moving averages. <em>Management science</em>, <em>6</em>(3), 324–342.</p><p>[5] Ariton, L. (2021). A Thorough Introduction to Holt-Winters Forecasting. <a href="https://medium.com/analytics-vidhya/a-thorough-introduction-to-holt-winters-forecasting-c21810b8c0e6">https://medium.com/analytics-vidhya/a-thorough-introduction-to-holt-winters-forecasting-c21810b8c0e6</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9a89f50d4ce1" width="1" height="1" alt=""><hr><p><a href="https://medium.com/amdax-asset-management/forecasting-the-bitcoin-price-using-holt-winters-exponential-smoothing-9a89f50d4ce1">Forecasting the Bitcoin price using Holt-Winters Exponential Smoothing</a> was originally published in <a href="https://medium.com/amdax-asset-management">Amdax Asset Management</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Stock-to-Flow: a philosophy, not a model]]></title>
            <link>https://medium.com/amdax-asset-management/stock-to-flow-a-philosophy-not-a-model-7d22c2eb7fd0?source=rss-865ada47c174------2</link>
            <guid isPermaLink="false">https://medium.com/p/7d22c2eb7fd0</guid>
            <category><![CDATA[bitcoin]]></category>
            <category><![CDATA[stock-to-flow]]></category>
            <category><![CDATA[econometrics]]></category>
            <category><![CDATA[quantitative-research]]></category>
            <dc:creator><![CDATA[Tim Stolte]]></dc:creator>
            <pubDate>Mon, 02 May 2022 14:50:10 GMT</pubDate>
            <atom:updated>2022-08-03T12:57:20.455Z</atom:updated>
            <content:encoded><![CDATA[<p>Ever since the Stock-to-Flow model was <a href="https://medium.com/@100trillionUSD/modeling-bitcoins-value-with-scarcity-91fa0fc03e25">introduced</a> by <a href="https://medium.com/u/bcb63a182704">PlanB</a> back in 2019, there has been quite some controversy on its validity. Those in favour put forward the well-founded scarcity argument and point at the historical accuracy of the estimates. Meanwhile, opponents argue that the underlying assumptions of the model do not adhere to econometric standards. For a quick overview of most points of critique on the model, I refer to <a href="https://ercwl.medium.com/a-list-of-the-greatest-blows-to-the-s2f-model-622b266ce199">this article</a> by <a href="https://medium.com/u/bdd9bfe4b591">Eric Wall</a>.</p><p>In a recent <a href="https://bitcoinmagazine.com/markets/whats-up-with-bitcoin-stock-to-flow">article</a> in Bitcoin Magazine, an attempt was made to refute the arguments against the Stock-to-Flow model. The author, <a href="https://medium.com/u/8a70b212d583">Harold Christopher Burger</a>, summarized some alleged <em>death blows</em> and made an effort to explain why we should not consider them as such. Sadly, he did not succeed. Let’s take a closer look at two of his arguments.</p><h3><strong>“Only large changes in stock-to-flow are meaningful”</strong></h3><p>The Stock-to-Flow model is autocorrelated. That is beyond dispute for both supporters and opponents. This means that the errors terms (i.e. the difference between the estimated and the actual price) are dependent on previous values and one should look at the <em>changes</em> in both the stock-to-flow ratio and price instead of the raw values themselves. However, we find that changes in stock-to-flow have no significant effect on the changes in price.<a href="#_ftn1">[1]</a> The author then goes on to conclude that this does not matter, as large shocks and small shocks may affect the price differently. He poses the question of whether we must really expect that both small and large changes in stock-to-flow cause changes in price. The answer is yes. We must really expect that, and for one simple reason. The core assumption of the Stock-to-Flow model specification is that the independent variable (in our case the ratio of the circulating supply and the issuance) LINEARLY affects the dependent variable (in our case the price). If you assume otherwise, the model specification should reflect this and you should account for this non-linearity. Hence, this result actually invalidates the Stock-to-Flow model as it is currently specified.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/619/1*lDNgQ0XE9qvmuN3ARSRC9A.png" /><figcaption>Stock-to-Flow model specification</figcaption></figure><p>Could we then alter the equation to distinguish between the effects of large and small shocks in the stock-to-flow ratio? Sure. In fact, that is exactly what <a href="http://www.kripfganz.de/">Sebastian Kripfganz</a> did in his <a href="https://www.youtube.com/watch?v=u_VkwA5fkLA">research</a> by considering the mining reward as a separate parameter in the model. This mining reward is the sole driver behind the large shocks that take place during every bitcoin halving. He documents no statistically significant effect of this reward series in his equations, leading us to conclude that there is no evidence to support the statement that only large shocks in stock-to-flow are relevant.</p><p><a href="#_ftnref1">[1]</a> For the sake of brevity, these results are included in the Appendix.</p><h3>“Cointegration is not a big thing”</h3><p>Just from looking at the graphs, it is easy to see that both the stock-to-flow and price series do not have a long-run unconditional mean and variance and these statistical properties change over time. In statistics, we are then talking about non-stationarity. Non-stationary time series are characterised by persistent shocks. This means that such series do not necessarily return to their original level in case of a shock. According to econometric theory, one can only extract valuable information from a one-factor linear regression if both metrics are stationary or if both metrics are non-stationary AND there exists cointegration between the two. Cointegration is a complicated and abstract term. A short non-technical explanation for it would be the following: two non-stationary series are cointegrated if and only if, to some degree, at least one of the two series shows a desire to return to the other one. For a more detailed and intuitive illustration, one can consult the following <a href="https://www.researchgate.net/publication/254330798_A_Drunk_and_Her_Dog_An_Illustration_of_Cointegration_and_Error_Correction">piece</a>.</p><p>Many sources (once again referring to <a href="https://ercwl.medium.com/a-list-of-the-greatest-blows-to-the-s2f-model-622b266ce199">this article</a>) have proven that cointegration does not hold in the Stock-to-Flow model. Nevertheless, in the Bitcoin Magazine it is stated that that problem is overblown. Why? Well, firstly because cointegration is a concept that is not used outside of econometrics. Is that a valid argument? No, since linear regression models such as the Stock-to-Flow model are purely econometric. It should therefore always rely on the fundamental basics of econometrics. Secondly, the author calls upon the Turing Award-winning computer scientist <a href="https://en.wikipedia.org/wiki/Judea_Pearl">Judea Pearl</a>, who has claimed to have never heard of cointegration and has yet to find someone who can satisfactorily explain it to him. Now, I am not questioning the integrity and/or brilliance of an individual that is definitely smarter than myself, but saying that something is irrelevant just because a bright mind says so should definitely be deemed a fallacy. Especially because an entire field of study would consider the absence of cointegration a death blow to a model.</p><h3>Conclusion: not a model</h3><p>Being an econometrician by profession, I am compelled to reject the validity of the Stock-to-Flow specification. Admittedly, it is easy for me to discredit someone’s hard work and not come up with an alternative myself. <a href="https://medium.com/u/bdd9bfe4b591">Eric Wall</a> gracefully avoided becoming the community’s buzzkill by <a href="https://twitter.com/ercwl/status/1250537262800142336">advocating</a> the rainbow chart after destroying Stock-to-Flow. I am not going to do that.</p><p>In all honesty, I believe that the pursuit for a long-term price model is pointless. There is no holy grail that can more or less accurately predict the bitcoin price for the coming months, let alone for the coming years. However, this does not mean that we cannot make predictions. The most careful among us will simply state that the best price prediction at any point in the future is the price of today. Others swear by the rainbow chart predictions, which are purely based on the assumption that we can generalise the historical growth rate of the price series and use it to predict future prices.</p><p>In the end, the rainbow model is not a valid linear regression on a causal variable, because time itself does not affect supply or demand. It is not a model, it is a philosophy. And whether you are willing to make the underlying assumptions determines whether you trust or distrust its predictions. The same goes for Stock-to-Flow. It may provide a reasonable guess of future price levels, but there is just no way of knowing how reasonable the guess is. For that you need statistics. Correct statistics.</p><h3>Appendix</h3><p>In my stock-to-flow analysis, I obtain daily price and stock-to-flow time series data from Glassnode. My sample period ranges from 1 August 2010 until 30 April 2022, yielding 4291 observations. I recognise that the original article considers monthly data from January 2009 until February 2019 and uses bitcoin’s total market value instead of price, but I assume that my choice of data will not result in any loss of generality. To those less familiar with statistics, you will find some definitions of all statistical terms used in this Appendix.</p><p>For the sake of replication, I estimate the following equation:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/362/1*hWeVQ9MH2-Aa-XOvx5zCfg.gif" /></figure><p>in which denotes the bitcoin price and denotes the stock-to-flow series. OLS estimation yields a coefficient β = 2.959. This coefficient, representing the effect of stock-to-flow on the price, is statistically significant at a 1% significance level. The R² of the estimation equals 0.914, which is more or less in line with the results of the original Stock-to-Flow article. The fitted values of the equation are plotted against the price in Figure 1. I would like to stress once more that no one should use this result to draw the conclusion that the price can be explained and/or predicted directly by the stock-to-flow metric. This equation cannot be considered an adequate econometric model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*D37uoGwm98zpL_GrsppJJA.png" /></figure><p>For the analysis of a model that would not violate the necessary assumption of either stationary series or cointegrated non-stationary series, we consider the following equation:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/401/1*gA6r_ba0N-AEsBrwmpCf8w.gif" /></figure><p>Note that this equation differs from the previous one as we have taken the first difference of both the price and stock-to-flow series. After estimating the model, we find β = 0.012 with a corresponding <em>p</em>-value of 0.062. This <em>p</em>-value means that we do not reject the hypothesis that the changes in the stock-to-flow metric do not affect the changes in price. To use a little less technical terms, this <em>p</em>-value confirms that we cannot state that the changes in stock-to-flow have any effect on price changes. This is also supported by the R² of 0.001, implying that stock-to-flow shocks would only explain around 0.1% of the total variation in price differences over the full sample period. Figure 2 provides more insights into what such an R² looks like. The fitted values in green (practically the zero line) only cover the smallest of price changes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*RqA4BTNv5dVf7k8U6qrk_A.png" /></figure><p>As promised, a small overview of all statistical terms and their explanations:</p><ul><li>OLS; short for Ordinary Least Squares, a method that is used to estimate the coefficients in a linear regression.</li><li>R²; the proportion of variation in the dependent variable (price) that is explained by the explanatory variable (stock-to-flow).</li><li>β; a model coefficient that measures the effect of one variable (in our case the stock-to-flow ratio) on another (in our case the price).</li><li><em>p</em>-value; based on an estimated model coefficient (for instance β) and measures the likeliness of the “true coefficient” being equal to zero.</li><li>Significance level; how likely it must be for the “true coefficient” to be equal to zero before we assume otherwise, usually 5% (<em>p</em>-value of 0.05). If <em>p</em>-value &lt; 0.05, we conclude that there is a significant effect.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7d22c2eb7fd0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/amdax-asset-management/stock-to-flow-a-philosophy-not-a-model-7d22c2eb7fd0">Stock-to-Flow: a philosophy, not a model</a> was originally published in <a href="https://medium.com/amdax-asset-management">Amdax Asset Management</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>