<?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 Oren Tapiero on Medium]]></title>
        <description><![CDATA[Stories by Oren Tapiero on Medium]]></description>
        <link>https://medium.com/@oren.j.tapiero?source=rss-e022478233c9------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*preKfxE2UeUX4YW6yqMjpg.jpeg</url>
            <title>Stories by Oren Tapiero on Medium</title>
            <link>https://medium.com/@oren.j.tapiero?source=rss-e022478233c9------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 04:56:13 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@oren.j.tapiero/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[De Gaulle: The Man Who Had to Liberate France Twice]]></title>
            <link>https://medium.com/@oren.j.tapiero/de-gaulle-the-man-who-had-to-liberate-france-twice-e98c0a5077fd?source=rss-e022478233c9------2</link>
            <guid isPermaLink="false">https://medium.com/p/e98c0a5077fd</guid>
            <category><![CDATA[france]]></category>
            <category><![CDATA[ww2]]></category>
            <category><![CDATA[history]]></category>
            <category><![CDATA[movies]]></category>
            <category><![CDATA[politics]]></category>
            <dc:creator><![CDATA[Oren Tapiero]]></dc:creator>
            <pubDate>Sat, 14 Nov 2026 18:13:49 GMT</pubDate>
            <atom:updated>2026-11-14T18:13:49.961Z</atom:updated>
            <content:encoded><![CDATA[<h3>Antonin Baudry’s two-part epic is not merely a story about resistance to Nazi Germany. It is a story about sovereignty, political legitimacy, and the limits of American power.</h3><p>There is something compelling about the timing of Antonin Baudry’s two films, <em>Résistance</em> and <em>Liberté</em>, arriving in cinemas amid the political and economic crisis currently unfolding in France.</p><p>Both movies take place during one of the darkest periods in French history. A period marked by a humiliating military defeat, shameful collaboration with Nazi Germany, the deportation of Jews to Nazi death camps, and the complete loss of French sovereignty and independence.</p><p>Yet these movies are not simply about World War II. They tell us something about France itself, its political history, and perhaps even the crisis facing Western democracies today.</p><p>Having grown up in France, I found the films particularly interesting. Despite my French upbringing, I was more familiar with the British and American narratives of World War II than with the French one.</p><p>Of course, French cinema and literature have dealt extensively with the occupation, collaboration, and Resistance. But what makes these two movies particularly interesting is their focus on a different story: De Gaulle’s struggle to preserve France as an independent, sovereign nation.</p><p>And that struggle was not directed exclusively against Nazi Germany.</p><h3>France: A Republic, Not an Ethnic Nation</h3><p>Although the present is far removed from the events of World War II, the movies seem particularly relevant amid France’s current political crisis and the growing tensions between parts of French society and communities descended from immigrants from Africa and the Maghreb.</p><p>They remind viewers that France cannot simply be understood as yet another European ethnic nation. Rather, it is a nation built around a set of shared political values: those of <em>La République</em>.</p><p>Several scenes beautifully illustrate this point.</p><p>Throughout both movies, we see the Free French forces composed of soldiers of different religions, ethnicities, and geographical origins.</p><p>One particularly moving scene takes place during the Battle of Bir Hakeim, where General Pierre Kœnig allows Jewish soldiers to mourn a fallen comrade according to their religious traditions. The soldiers recite the Kaddish, the Jewish prayer for the dead, in the middle of the desert.</p><p>It is a powerful scene, almost as if the director wanted to remind today’s viewers that freedom of religion is a foundation of French republican values, and that Jews have made important contributions to French history.</p><p>Indeed, several prominent Jewish intellectuals played important roles in the Free French movement in London. Among them were René Cassin, one of De Gaulle’s closest collaborators, and the celebrated author Albert Cohen, who worked in London as a representative of the Jewish Agency and maintained close contacts with Free France.</p><p>Nor should we forget the important roles played by Jews and Armenians in the French Resistance against Nazi Germany. Robert Guédiguian’s 2009 film <em>L’Armée du crime</em> (<em>The Army of Crime</em>) powerfully portrays their contributions, telling the story of the Manouchian Group, led by the Armenian poet and resistance fighter, Missak Manouchian. The group brought together Jewish, Armenian, and other immigrant fighters who risked their lives to liberate France, a country they considered their own and whose republican values they were determined to defend.</p><p>Another scene shows fighters from North Africa, serving under General Leclerc, overwhelming Rommel’s forces.</p><p>In fact, a significant part of both movies takes place in what were then French colonies in Africa.</p><p>This is hardly accidental. It reminds contemporary viewers of France’s historical connections with Africa and the Maghreb, and of the contribution soldiers from these territories made to France’s liberation.</p><p>It also offers a different perspective on today’s debates about French national identity and immigration.</p><h3>Political Legitimacy: Must Great Leaders Be Elected?</h3><p>The movies also tell us something fascinating about French political history.</p><p><strong>Some of France’s most transformative historical figures emerged from crises rather than from established democratic or political institutions.</strong></p><p>Consider Joan of Arc, Napoleon Bonaparte, and Charles de Gaulle.</p><p>None initially derived their authority from an electoral mandate. Their legitimacy emerged from extraordinary historical circumstances and from their ability to embody a particular vision of France.</p><p>De Gaulle is perhaps the most striking example.</p><p>When he arrived in London in June 1940, he had neither an electoral mandate nor a political party behind him. He commanded relatively few soldiers and had very limited resources.</p><p>Yet he claimed to represent France.</p><p>This lack of formal legitimacy becomes a central political obstacle throughout the movies, particularly in his relationship with U.S. President Franklin D. Roosevelt.</p><p>One of the most interesting aspects of the films is how they portray De Gaulle’s remarkable political cunning. Despite his weak position, he gradually secures recognition from leaders commanding vastly superior military and economic resources.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/620/1*1Bb8m_G4zQYxhOv5QUfJvg.jpeg" /><figcaption>De Gaulle (Simon Abkarian) “snatching” formal recognition from U.S. President Roosevelt (Campbell Scott)</figcaption></figure><p>Those familiar with French history will remember that something similar happened in 1958.</p><p>De Gaulle returned to power without a major political party behind him, although this time through constitutional procedures and with parliamentary approval. He subsequently became president of the newly established Fifth Republic.</p><p>The parallel with contemporary France is difficult to ignore.</p><p>Today, neither Marine Le Pen on the far right nor Jean-Luc Mélenchon on the far left seems capable of inspiring much hope for better days.</p><p>It is almost as if the movies are suggesting that genuine political change will not emerge from within the existing political system, but from outside it.</p><p>And this observation extends well beyond France.</p><p>As Western democracies face profound crises of political legitimacy and public confidence, the need for change is increasingly apparent. Yet it is difficult to imagine that change emerging from the same political institutions and parties that seem incapable of addressing these crises.</p><h3>De Gaulle’s Two Battles: Against Germany and the United States</h3><p>Perhaps the most fascinating aspect of these movies is how they interpret France’s struggle for independence.</p><p>The French narrative of World War II is deeply complicated.</p><p>On the one hand, there is the shameful defeat and collaboration of a nation that considered itself among the greatest in the world.</p><p>On the other hand, a relatively small group of truly patriotic Frenchmen left their mark on history despite very limited resources.</p><p>But the movies introduce another dimension.</p><p><strong>De Gaulle was fighting two very different battles: one against Nazi Germany, and another against the prospect of France becoming politically subordinate to its American liberators.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9xCov2TrbVtbekDqgzhsKQ.png" /><figcaption><strong>De Gaulle (Simon Abkarian) and Henri Giraud (Thieety Lhermitte).</strong> There is something symbolic about Giraud’s cap, which resembles the one worn by Marshal Philippe Pétain, as if Giraud were yet another Pétain, only this time serving American interests.</figcaption></figure><p>Interestingly, the Germans remain largely in the background of these movies. We barely see them, except during the major military confrontations.</p><p>By contrast, the Americans occupy a much more prominent place in the story.</p><p>And unlike the conventional Anglo-American narrative of World War II, in which the United States appears primarily as the liberator of France, these movies portray American policy as a serious threat to French sovereignty and independence.</p><p>One scene shows the U.S. Army training officers in French so they could assume the roles of prefects and administer France after its liberation.</p><p>An even more unsettling scene shows Roosevelt ordering the removal of any mention of the French Republic from what are supposed to be French banknotes, leaving only the French flag.</p><p>These scenes refer to the Allied Military Government of Occupied Territories (AMGOT) arrangements, through which the Allies prepared to administer liberated territories.</p><p>The movies attribute considerable importance to these plans, far more than they generally receive in contemporary British and American narratives of the war.</p><p>To be clear, Nazi Germany and the United States represented fundamentally different threats. The Germans occupied France and sought to dominate it by force, whereas the Americans were fighting to liberate Europe from Nazi rule.</p><p>Nevertheless, Roosevelt’s reluctance to recognize De Gaulle and his desire to retain Allied authority over the liberation posed a serious political challenge to the Free French leadership.</p><p>It is almost as if the movies are suggesting that <strong>De Gaulle was fighting not only to liberate France from Nazi occupation, but also to ensure that liberation would not come at the expense of French independence.</strong></p><p>In this sense, France had to be liberated twice.</p><p>First from Nazi Germany, and then from the possibility of political subordination to the United States.</p><h3>Churchill, Roosevelt, and the Origins of Gaullism</h3><p>Seen from this perspective, De Gaulle’s foreign policy after 1958 becomes much easier to understand.</p><p>His insistence on French independence from American influence was not merely a Cold War political position. It was rooted in his wartime experience.</p><p>This helps explain his often difficult relationship with Washington, his determination to develop an independent French nuclear deterrent, and his decision to withdraw France from NATO’s integrated military command in 1966.</p><p>The movies also portray the relationship between De Gaulle and British Prime Minister Winston Churchill beautifully.</p><p>It is a relationship characterized by deep mutual respect, but also by considerable suspicion.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8DpzNtiHYbA2qeZ_q-ZYcQ.jpeg" /><figcaption><em>Churchill (Simon Russell Beale) and De Gaulle (Simon Abkarian) have dinner at the beginning of World War II. The setting is romantic, with candles and a bottle of champagne, yet the atmosphere is thick with mutual suspicion.</em></figcaption></figure><p>Churchill clearly admires De Gaulle’s determination and courage. Yet when forced to choose between his French ally and his American ally, Churchill repeatedly prioritizes the latter.</p><p>De Gaulle understands this perfectly well.</p><p>And the tension between the two men would resonate long after the war.</p><p>It helps explain De Gaulle’s subsequent opposition to British membership in the European Economic Community, the predecessor of today’s European Union.</p><p>From De Gaulle’s perspective, Britain was simply too closely aligned with the United States to participate fully in his vision of an independent Europe.</p><p>The roots of Gaullism, particularly its emphasis on national sovereignty and strategic independence, are therefore clearly visible throughout these films.</p><h3>Can Democracies Still Produce Leaders Like De Gaulle?</h3><p>Perhaps the most unsettling message of these two films has little to do with World War II itself.</p><p>De Gaulle emerged when the French political establishment had collapsed, when the country’s institutions had lost legitimacy, and when foreign powers were deciding its future.</p><p>He possessed neither an electoral mandate nor a powerful political party.</p><p>Yet he managed to impose his vision of France on leaders commanding armies and resources incomparably greater than his own.</p><p>This story remains profoundly relevant today.</p><p>Across Western democracies, political institutions increasingly struggle to inspire confidence, while established political parties seem incapable of producing leaders with a convincing vision for the future.</p><p>But there is also a paradox here.</p><p>De Gaulle emerged from outside the established political system, yet he eventually created the Fifth Republic, a political system built around strong presidential leadership.</p><p>Today, that very system appears increasingly incapable of producing the kind of transformative leadership that De Gaulle himself embodied.</p><p>Of course, there is nothing inherently democratic about a leader who claims to embody the nation without first receiving an electoral mandate. History offers plenty of examples of where such claims can lead.</p><p>But De Gaulle’s story raises a deeper question about the relationship between democratic legitimacy and political leadership.</p><p><strong>Perhaps the real question these films raise is not whether another De Gaulle will emerge. It is whether our democratic institutions can still produce transformative leadership without first experiencing the kind of catastrophe that made De Gaulle possible.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e98c0a5077fd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Modelling crypto order-flow imbalance as an additive–multiplicative process — or, detecting whale…]]></title>
            <link>https://medium.com/@oren.j.tapiero/modelling-crypto-order-flow-imbalance-as-an-additive-multiplicative-process-or-detecting-whale-fc7a6bd858a9?source=rss-e022478233c9------2</link>
            <guid isPermaLink="false">https://medium.com/p/fc7a6bd858a9</guid>
            <category><![CDATA[market-microstructure]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[algorithmic-trading]]></category>
            <category><![CDATA[quantitative-finance]]></category>
            <category><![CDATA[crypto-trading]]></category>
            <dc:creator><![CDATA[Oren Tapiero]]></dc:creator>
            <pubDate>Fri, 08 May 2026 11:21:44 GMT</pubDate>
            <atom:updated>2026-05-08T11:21:44.065Z</atom:updated>
            <content:encoded><![CDATA[<h3>Modelling crypto order-flow imbalance as an additive–multiplicative process — or, detecting whale activity (well… sort of)</h3><p><em>Why concentrated, self-amplifying flow doesn’t always move prices — and how to tell when it will.</em></p><h3>The setup</h3><p>Most order-flow imbalance (OFI) models in crypto treat imbalance as an <em>explanatory variable</em> for returns. You regress next-bar returns on signed taker-buy share, get a price-impact coefficient, call it a day. The imbalance itself is just an input.</p><p>This piece argues for a different lens: <strong>OFI should be viewed as a stochastic process that merits its own modeling.</strong> When modeled accurately, it naturally yields three insights: heavy-tailed returns, a continuous real-time assessment of whether the market is <strong>driven by self-amplifying flow or background noise</strong>, and a partial answer to a question every crypto practitioner eventually asks: <em>are the whales doing something right now?</em></p><p>The phrase “well… sort of” in the title really captures the essence here. This framework isn’t about tracking individual whales or pinpointing which wallet is making moves. Instead, it focuses on identifying the distinct <strong>microstructural pattern</strong> of the flow that whales usually create — think of it as concentrated, directional, and self-amplifying. It does this continuously and in real time using two OLS regressions, without depending on any on-chain data. Whether you see this as a “whale detection” tool really hinges on how you define the term. As an indicator of whale-like flow patterns, it makes sense. However, if you’re searching for a tool to identify specific wallets, this isn’t the one.</p><p>The framework is an <strong>additive × multiplicative</strong> stochastic process. The empirical work covers BTC, ETH, BNB, SOL, and XRP at 4-hour frequency, 2019–2024, on Binance.</p><p>The headline result, before any of the math: <strong>state-dependent OFI amplification is a persistent structural feature of crypto order flow</strong> — not a crisis artifact. Across all five assets and twelve independent six-month sub-periods, the multiplicative variance component is positive and significant, with Fama–MacBeth t-statistics between 3.26 (BTC) and 5.05 (XRP).</p><p>But persistence at the <em>process</em> level doesn’t mean persistence at the <em>price</em> level. The transmission from OFI dynamics to return variance is gated by market depth, and that gate has been mostly closed since late 2022. That’s the more interesting story, and we’ll get to it.</p><h3>Why additive×multiplicative</h3><p>Take the relative OFI:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/248/1*z3WdYcxK7_Co_JbrPDD1Ig.png" /></figure><p>A value near +1 means buy-initiated flow dominates; near −1, sell-initiated. This is just normalised signed taker volume, nothing fancy.</p><p>Now model its dynamics as:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/303/1*vEOTLydb5Am6EzWtAQaRAQ.png" /></figure><p>Two independent Wiener processes. One scales with the current state (the multiplicative term, with diffusion coefficient sigma_m times I_t), one doesn’t (the additive term, sigma_a). The drift is mean-reverting , nu is negative across all five assets and all twelve sub-periods.</p><p>The conditional variance of OFI changes is:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/230/1*MWt5vUw_kMFT728H1zyXkA.png" /></figure><p>This is the whole engine. When the imbalance is small, the additive piece dominates and the process behaves like a vanilla noise-driven system. When the imbalance is large, the multiplicative piece kicks in: extreme states beget more extreme states. State-dependent amplification.</p><p>Discretised, this becomes a Kesten recurrence with power-law stationary tails — heavy tails emerge as a structural consequence of the mechanism, not as a distributional assumption.</p><p><strong>What the multiplicative noise actually represents.</strong> Three participant types, all well-documented:</p><ul><li><strong>Whales</strong> — large holders whose position sizes produce flow disproportionate to routine trading noise (Bouchaud et al. 2009; Makarov &amp; Schoar 2020). When a whale moves, the size of their order is set by their book, not by current market state — but the <em>price response</em> to that order scales with where the market already is. That’s multiplicative.</li><li><strong>Trend followers</strong> who reinforce the prevailing direction (De Long et al. 1990; Cutler et al. 1990). Their flow is explicitly a function of recent state.</li><li><strong>Leveraged traders</strong> who unwind into falling markets, generating self-reinforcing sell-side pressure (Geanakoplos 2010; Brunnermeier &amp; Pedersen 2008). Margin calls are state-triggered by construction.</li></ul><p>When any of these dominate, current imbalance state determines future imbalance variance. That’s what a positive multiplicative variance component is measuring. The model can’t tell you which of the three is active at any moment, it just tells you that <em>something</em> state-dependent is driving flow. In crypto, given on-chain ownership concentration, the whale channel is usually a reasonable first guess.</p><h3>Estimation: a two-step OLS that anyone can run</h3><p>The estimation is embarrassingly simple. First, the conditional mean:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/146/1*XQey4pMzUqRckSzis2LLlQ.png" /></figure><p>Then the conditional variance:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/156/1*-JLV0mwgtxa1JynVgdbsvg.png" /></figure><p>Two OLS regressions. The intercept of the second one is the additive variance; the slope on the squared imbalance is the multiplicative variance. Clip negatives at zero (variances must be non-negative) and you’re done.</p><p>In about twenty lines of NumPy:</p><pre>import numpy as np<br><br>def estimate_am_ofi(I, clip_negative=True):<br>    &quot;&quot;&quot;<br>    Two-step OLS estimator for the additive-multiplicative OFI model.<br><br>    Parameters<br>    ----------<br>    I : array-like<br>        Relative OFI series in [-1, 1], shape (T,).<br><br>    Returns<br>    -------<br>    dict with keys: nu, sigma2_a, sigma2_m, threshold, s_m<br>    &quot;&quot;&quot;<br>    I = np.asarray(I, dtype=float)<br>    dI = np.diff(I)<br>    I_lag = I[:-1]<br><br>    # Step 1: conditional mean — recover the drift parameter<br>    nu = np.cov(dI, I_lag, ddof=1)[0, 1] / np.var(I_lag, ddof=1)<br>    u = dI - nu * I_lag<br><br>    # Step 2: conditional variance — regress squared residuals on squared OFI<br>    X = np.column_stack([np.ones_like(I_lag), I_lag**2])<br>    coef, *_ = np.linalg.lstsq(X, u**2, rcond=None)<br>    sigma2_a, sigma2_m = coef<br><br>    if clip_negative:<br>        sigma2_a = max(sigma2_a, 0.0)<br>        sigma2_m = max(sigma2_m, 0.0)<br><br>    # Dominance threshold and per-bar multiplicative share<br>    threshold = np.sqrt(sigma2_a / sigma2_m) if sigma2_m &gt; 0 else np.nan<br>    s_m = (sigma2_m * I_lag**2) / (sigma2_a + sigma2_m * I_lag**2 + 1e-12)<br><br>    return {&quot;nu&quot;: nu, &quot;sigma2_a&quot;: sigma2_a, &quot;sigma2_m&quot;: sigma2_m,<br>            &quot;threshold&quot;: threshold, &quot;s_m&quot;: s_m}</pre><p>That’s the whole estimator. Feed it a relative OFI series and it returns the drift, both variance components, the threshold, and the per-bar multiplicative share.</p><p>To check robustness, split the sample into K = 12 six-month sub-periods, run the estimator within each, and compute Fama–MacBeth t-statistics. Across all five assets:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1yR_NaXipn1GvvtOPjRkoQ.png" /><figcaption><em>Fama–MacBeth estimates across K = 12 six-month sub-periods. The multiplicative variance component σ̂ₘ² is positive and significant for every asset, with t-statistics between 3.26 and 5.05. The dominance threshold σ̂ₐ/σ̂ₘ is the imbalance level above which multiplicative variance dominates additive variance — well-identified for all five assets.</em></figcaption></figure><p>The multiplicative component is significant for every asset across every sub-period. Mean reversion holds everywhere. And the dominance threshold — the imbalance level at which multiplicative variance equals additive variance — is well-identified and economically interpretable.</p><p>That threshold is the most useful single number the model produces. It tells you, asset by asset, how much directional concentration the market can absorb before self-amplifying dynamics take over.</p><p>The cross-sectional ordering is clean: ETH (0.430) &gt; SOL (0.418) &gt; BTC (0.328) &gt; BNB (0.277) &gt; XRP (0.227). Deeper books require larger imbalances; concentrated-ownership assets like XRP flip into multiplicative dominance with relatively modest flow.</p><h3>The multiplicative share as a regime indicator</h3><p>Define the share of conditional OFI variance attributable to the multiplicative channel:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/137/1*nayAroJb1lLIwYizJeZz2Q.png" /></figure><p>This is bounded in [0, 1]. Smooth it with a 10-bar moving average and you get a continuous regime indicator: how much of right-now’s OFI variance is being driven by state-dependent amplification versus background noise.</p><p>Empirically, the multiplicative share is compressed near zero most of the time, punctuated by sharp episodic spikes. The 80th-percentile threshold sits around 0.07 for BTC and 0.05 for ETH. Above the threshold, the regime is active; below it, the additive baseline dominates.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AcNGgh2-bh1cnpWhu_UKxA.png" /><figcaption><em>Figure 1. BTCUSDT price (log scale, top) and smoothed multiplicative share s_m (10-bar mean, bottom) over 2019–2024. Dashed line marks the 80th-percentile threshold. Vertical dashed lines mark the eight annotated market events. The episodic structure is the point: long quiet stretches punctuated by sharp spikes that align with COVID, the May 2021 sell-off, LUNA, and other crisis events.</em></figcaption></figure><p>For BTC, four episodes carve up the sample:</p><ol><li><strong>Thin liquidity era (mid-2019 to early 2020).</strong> Sustained moderate elevation, peak 0.62. Choppy, no single shock.</li><li><strong>COVID crash (Feb–May 2020).</strong> Local peak Feb 3, <em>37 days</em> before Black Thursday’s ~50% collapse. Pre-crash repositioning produced concentrated directional flow that resolved into the crash later.</li><li><strong>FTX collapse (Oct 2022 — Feb 2023).</strong> Peak Jan 10, 2023 — <em>two months after</em> FTX failed. The peak coincides with the recovery rally, not the collapse. The collapse itself was a forced-liquidation event too fast for the OFI channel to build a multiplicative signal beforehand.</li><li><strong>Institutional bull run (2024).</strong> 26.9% of bars above threshold over the year, a sustained baseline rather than a sharp spike. ETF-driven concentrated flow without the acute stress of crisis episodes.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QzknyDrL_Po41E8mfMl2Qg.png" /><figcaption><em>Figure 2. BTCUSDT s_m across the four episodes. Each column is one episode, with price on top and the multiplicative share on bottom. The qualitative differences are the whole point: the COVID spike (column 2) is shaped completely differently from the 2024 sustained baseline (column 4). One is acute stress; the other is structural concentration without crisis.</em></figcaption></figure><p>Episode 4 is the structural anomaly. The OFI process is in a multiplicative regime almost continuously, yet realised return variance stays muted and the path remains orderly. To understand why, we need the price-impact equation.</p><h3>From OFI variance to return variance — the structural hinge</h3><p>This is where the framework earns its keep. Everything up to now has been about the OFI process in isolation. The question that matters for trading is whether any of it transmits to prices, and if so, when.</p><p>The bridge is the contemporaneous price-impact equation:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/163/1*myDYAhRYWPcSLtzKiDRK7g.png" /></figure><p>A familiar object, but here it’s doing structural work, not just regression work. It’s the operator that maps an OFI state into a return. Once you have it, you can take the conditional variance of next-period returns <em>given the current imbalance state</em>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/295/1*M9y_U6uQsZFtdc83mQNZow.png" /></figure><p>And the inner term, the conditional variance of next-period OFI given the current state, is exactly what the additive multiplicative model gives us. Substitute, and you get the central structural result:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/290/1*FkKJblTKtvUJs7vcmES4uw.png" /></figure><p>Read this slowly. The conditional variance of returns inherits the additive multiplicative structure of OFI, scaled uniformly by the squared price-impact coefficient. Three terms, doing three different jobs:</p><p>The <strong>first term is</strong> a state-independent baseline. Always there. This is the return variance you’d get if OFI were a pure additive, Gaussian-like process.</p><p>The <strong>second term</strong> is the state-dependent return variance. This is the term that lights up when the imbalance is large, and it’s the entire reason the framework exists. Concentrated flow today produces more variable returns tomorrow, with the magnitude scaling quadratically in the imbalance.</p><p>The <strong>third term is</strong> the residual return variance not attributable to OFI. Information shocks, news, everything else.</p><p>The fraction of forecast return variance coming from the multiplicative channel is the thing you can actually compute on live data:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/219/1*ufHfVa8SHlZiIVtrm3ysIg.png" /></figure><p>This is the whole risk-management story, compressed into one expression. Three things have to line up for a large return realisation.</p><p>First, the multiplicative variance parameter has to be non-trivial. The asset has to have state-dependent dynamics at all. Persistent across all five assets, established structurally.</p><p>Second, the imbalance has to be currently sitting in a multiplicative state. Time-varying, observable in real time.</p><p>Third, and this is the one that matters: the price-impact coefficient has to be open. Time-varying, regime-dependent, and crucially, not under the trader’s control.</p><p>That third condition is where everything interesting happens. Cont, Kukanov &amp; Stoikov (2014) gave us the inverse relationship between price impact and market depth: deep books give a small price-impact coefficient; thin books give a large one. So the same elevated multiplicative state produces wildly different return variance depending on whether the book is thick or thin. A fully loaded multiplicative OFI regime gets quarantined from prices when depth is sufficient. The flow is there. The amplification is there. But the price doesn’t move, because liquidity absorbs it.</p><p>This is the equation that makes the 2024 dissociation interpretable rather than mysterious. And it’s the equation that tells you why “concentrated flow detected” is not the same as “trade signal.” You need to verify, separately, that the price-impact channel is awake.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1014/1*gT0hBoh8Uh53r9dR__zV_Q.png" /><figcaption><em>Figure 3. Smoothed Amihud illiquidity ratio (orange, left axis) and smoothed absolute rolling beta (blue, right axis) for each of the five assets, 2019–2024. The co-movement across BTC, ETH, and SOL is the depth-absorption mechanism made visible: thinner markets, larger price-impact coefficients. BNB inverts — likely because Binance-specific flows confound the depth measurement on its own native token.</em></figcaption></figure><h3>The 2024 dissociation</h3><p>Estimate the price-impact coefficient as a 250-bar rolling regression of next-period returns on current OFI. For BTC, ETH, BNB, and SOL, the coefficient collapses toward zero after the FTX crisis and stays there throughout 2024.</p><p>The post-FTX compression is the most economically significant time-series feature in the data. It produces a striking pattern in 2024: the multiplicative share of OFI variance is persistently elevated, but the multiplicative share of return variance stays compressed near zero. The OFI process is multiplicative; the return variance channel is not.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/765/1*xChMamrdwlL5KW_IHq3Dtg.png" /><figcaption><em>Figure 4. BTCUSDT s_m,r across the same four episodes as Figure 2. The 2024 column collapses to near zero despite elevated s_m. Compare to the FTX column (third), where both channels are simultaneously active and s_m,r reaches its sample maximum. This is the dissociation, made visible.</em></figcaption></figure><p>What’s happening structurally: the spot Bitcoin ETFs approved in January 2024 brought large institutional flows that produced concentrated directional pressure, raising the multiplicative share of OFI variance. But those same institutions also brought deep liquidity provision, collapsing the price-impact coefficient. The flow was multiplicative; the depth absorbed it. Return path: +121% on the year, max drawdown −8.2%.<br>Compare to FTX. October 2022 to February 2023, both the multiplicative share and the squared price-impact coefficient are simultaneously large. The OFI channel explains 10.6% of realised squared returns in that episode, roughly ten times its explanatory power in any other episode. The multiplicative-to-additive ratio inside the OFI variance term hits 52.1, meaning the channel is almost entirely multiplicative-driven during this window.</p><p>This is the practitioner-grade synthesis: multiplicative OFI is necessary but not sufficient for large return realisations. You need concentrated flow and a thin book simultaneously. Either alone won’t do it.</p><h3>Crash-phase betas: the inversion</h3><p>Classify each bar by market phase using BTC dynamics: Bull, Bear, Crash, Recovery. The mean rolling price-impact coefficient within each phase reveals something quietly important:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*01XTkwCCUJKBgs_swR_kgg.png" /><figcaption><em>Mean rolling β̂ within crash-phase bars. BTC and ETH show negative crash-phase betas — buy-side flow predicts negative next-bar returns, the signature of informed sellers absorbing incoming liquidity. SOL inverts: a positive crash-phase beta of +137.9 reflects the LUNA–FTX period when the SOL sell side effectively disappeared and surviving buy-side flow had unbounded price impact.</em></figcaption></figure><p>For BTC and ETH, the crash phase beta inverts: buy side OFI predicts negative next bar returns. The mechanism is straightforward. Informed sellers absorb incoming buy initiated liquidity to unwind positions, then push prices down. The usual positive relationship between OFI and returns reverses.</p><p>SOL’s crash phase beta is the anomaly. Strongly positive. The microstructural signature of a market in existential stress: during the LUNA and FTX episodes, the sell side effectively disappeared, so any surviving buy side flow had unbounded price impact.</p><p>XRP is the other outlier, with a positive price impact coefficient across all phases. Momentum driven microstructure dominated by directional, trend following participants. This is also why XRP has the lowest dominance threshold (0.227). Concentrated ownership and an idiosyncratic regulatory backdrop mean the market flips into multiplicative dominance with very modest flow.</p><h3>What the model does NOT do</h3><p>This matters, especially if you’re going to use it for trading purposes.</p><p>The multiplicative share has leading indicator properties for specific crisis events. COVID showed a peak lag of −20 bars with peak correlation 0.36; LUNA showed −18 bars and 0.60. Both events featured a gradual buildup of concentrated directional flow that the model identified weeks before the price collapse.</p><p>But this does not generalise to an unconditional timing signal. Define a spike onset as the first bar of a contiguous run above the 70th percentile threshold, measure forward distance to the next 2σ price move, compare against a memoryless geometric null using Kolmogorov Smirnov. The geometric null is not rejected for any asset. The forward distance distribution is consistent with a memoryless process at the unconditional level.</p><p>Translation: the multiplicative share does not, on average across the full sample, produce statistically robust forward timing. The leading indicator behaviour is episode specific, concentrated in events with gradual flow buildup, and gets diluted by events triggered by discrete exogenous shocks (FTX insolvency, ETF approval, regulatory actions) where the regime activates in response rather than in anticipation.</p><p>So treat the multiplicative share as a structural state variable identifying periods of concentrated, self amplifying flow, not as a short horizon trigger. That’s the boundary the KS test draws, and it’s worth respecting.</p><h3>What you can do with this</h3><p>A few practical takeaways for someone hunting alpha in crypto microstructure.</p><p><strong>The dominance threshold is a meaningful, asset specific calibration constant.</strong> It tells you when the linear price impact assumption (Almgren Chriss style) is going to fail you. Above the threshold, expect non linearity, expect tail risk in execution, expect the price impact coefficient to be the wrong functional form.</p><p><strong>The two channel decomposition is operationally useful even without a forecast.</strong> The OFI multiplicative share and the return variance multiplicative share can agree or diverge. When they agree, you’re in a coherent regime. When they diverge, like 2024, the standard intuition that concentrated flow produces big moves breaks down. That divergence itself is information.</p><p><strong>Cross asset synchronisation is a parameter free systemic indicator.</strong> When three or more of five majors are simultaneously in their own multiplicative regime, you’re in a systemic episode (COVID and LUNA both showed sustained synchronisation above 0.60). When the spike is confined to one asset, it’s idiosyncratic (XRP’s legal resolution). No tuning required beyond the per asset threshold.</p><p><strong>The depth conditioning of price impact is the rate limiting step for any OFI based return forecast.</strong> If the price impact coefficient is compressed (which it has been for BTC since FTX), don’t bet on OFI driven return predictions. The channel is closed. Wait for the coefficient to wake up. It will, eventually, when depth thins.</p><h3>Where this goes next</h3><p>Three obvious extensions for anyone who wants to push this further:</p><p><strong>On-chain ownership concentration.</strong> The cross-sectional ordering of dominance thresholds correlates with intuitions about ownership concentration, but it isn’t tested formally. Crypto markets have something traditional asset classes don’t: observable wallet-level wealth distributions. Gini coefficients, Pareto exponents, large-holder activity ratios. Plug those in as cross-sectional explanatory variables for the threshold and you get a direct empirical link between on-chain structure and microstructural amplification.</p><p><strong>Better depth proxies.</strong> The Amihud (2002) illiquidity ratio is what was available at four-hour frequency for the full 2019–2024 sample. It’s a noisy proxy. Repeating the depth-absorption analysis with order-book-level resting liquidity (aggregate depth within five basis points of the midpoint) would turn the suggestive evidence into a clean test, and probably resolve the BNB anomaly, where the rolling correlation between illiquidity and the absolute price-impact coefficient inverts — likely because Binance-specific flows confound the depth measurement on its own native token.</p><p>Multivariate extension. The single-asset framework treats the five coins independently, then aggregates via a synchronisation indicator. A genuinely multivariate additive multiplicative model, with cross-asset multiplicative spillovers, would formalise the systemic-risk channel and probably reveal lead-lag structure across assets that the univariate work misses.</p><h3>Closing</h3><p>The additive×multiplicative lens does two things at once. It generates heavy-tailed OFI as a <em>consequence</em> of the participant structure rather than as an assumption, and it gives you a continuous, real-time decomposition of how much of right-now’s market is being driven by self-amplifying dynamics versus background noise.</p><p>So, is this whale detection? Sort of. It’s detection of <em>the kind of flow whales produce</em>, which is a strictly weaker claim but a useful one. You don’t get a wallet address out of it; you get a regime indicator that flips on when concentrated, state-dependent flow dominates the imbalance process, regardless of whether that flow comes from one whale, a coordinated group, a forced-liquidation cascade, or a wave of trend followers. For a practitioner trying to size positions or set execution thresholds in real time, that’s often the question that actually matters.</p><p>Whether those dynamics matter for <em>prices</em> is a separate question, gated by depth. Most of the time, in 2024, they didn’t. During FTX, they did, a lot.</p><p>The bigger lesson, the one I think generalises beyond crypto: when you have a microstructural process that can flip between additive and multiplicative regimes, modelling that flip explicitly is worth more than fitting one more GARCH variant to the returns it produces. The returns are downstream of the process. Model upstream.</p><p>Full paper: <a href="https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6688399">“Modelling Crypto Asset Order Flow Imbalance as an Additive and Multiplicative Process,”</a> April 2026.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fc7a6bd858a9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Avellaneda and Lee Is Not Dead. It Is Just Much Harder to Trade Than It Looks]]></title>
            <link>https://medium.com/@oren.j.tapiero/avellaneda-and-lee-is-not-dead-it-is-just-much-harder-to-trade-than-it-looks-b6dac83b5dfa?source=rss-e022478233c9------2</link>
            <guid isPermaLink="false">https://medium.com/p/b6dac83b5dfa</guid>
            <category><![CDATA[trading]]></category>
            <category><![CDATA[statistical-arbitrage]]></category>
            <category><![CDATA[stock-trading]]></category>
            <category><![CDATA[quantitative-finance]]></category>
            <dc:creator><![CDATA[Oren Tapiero]]></dc:creator>
            <pubDate>Tue, 21 Apr 2026 09:45:55 GMT</pubDate>
            <atom:updated>2026-04-21T09:45:55.435Z</atom:updated>
            <content:encoded><![CDATA[<h3>A re-examination of one of quantitative finance’s most elegant statistical arbitrage frameworks and what survives once fees, hedging, and real holding periods enter the picture</h3><p>Avellaneda and Lee (2010) is one of those papers that stays with you.</p><p>It is elegant without being obscure. It takes a familiar intuition, that stocks can temporarily drift away from their fair relative value, and turns it into a disciplined statistical framework. The core move is simple. Decompose a stock’s return into a systematic component and an idiosyncratic residual, then model that residual as a mean reverting process. If the residual moves far enough away from equilibrium, bet on reversion.</p><p>As a research template, it is beautiful. As a trading strategy, the real question is much harsher: does it still make money?</p><p>That was the question I set out to answer by testing the framework on the S&amp;P 500 over 2015 to 2020. I examined six implementations: ETF-based and PCA-based residual engines, hedged and unhedged specifications, and a Ledoit-Wolf covariance variant. I also looked beyond the headline equity curves and broke the results down by sector, liquidity, market regime, trade anatomy, holding period, transaction cost sensitivity, and Fama-French factor attribution.</p><p>The headline result is blunt. Only one of the six baseline variants finished with a positive net return, and that result was so small that it was hard to distinguish from zero in any economically meaningful sense. The other five lost money after realistic costs. The best performer was the unhedged PCA variant, which earned just +0.1% net annualized with a net Sharpe of +0.04. The weakest, a single-ETF hedged version, lost about 3.6% net annualized.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Uluri0Icld7l3m6n6PMfHw.png" /></figure><p>At first glance, that sounds like the familiar ending to many academic strategies: elegant in theory, dead in implementation.</p><p>But that is not actually the right conclusion.</p><p>What I found is more interesting, and more useful. The Avellaneda and Lee framework does contain a real signal. The problem is that the signal is fragile, short-lived, and extremely sensitive to how it is executed. What fails is not the logic of residual mean reversion itself. What fails is the standard implementation once you layer on retail-like fees, hedge rebalancing, and holding periods that are too long for the signal’s actual half-life.</p><h3>The elegance of the original idea</h3><p>The framework begins with a very appealing premise. A stock’s return can be split into two parts: the portion explained by common factors and the residual portion that is specific to the stock. In the original spirit of the paper, those factors can be approximated either with related ETFs or with latent factors extracted from the cross-section through principal component analysis. The residual is then accumulated and modeled as an Ornstein-Uhlenbeck-like process, giving a standardized signal, the s-score, that tells you how far the asset has wandered from equilibrium.</p><p>That is a genuinely strong piece of financial engineering. It gives you a concrete answer to a hard trading problem: relative to what benchmark is this stock rich or cheap right now? And it gives you a trading rule that is coherent rather than impressionistic.</p><p>The attraction of the framework is that it feels like a proper bridge between asset pricing and execution. You are not merely saying that something looks stretched. You are defining stretch relative to a factor-implied fair level, then standardizing the deviation through a dynamic mean reversion model.</p><p>This is why the paper remains such a useful object of study even now. It teaches not only a strategy, but a way of thinking.</p><h3>What happens when you actually test it</h3><p>The six implementations I tested produced a remarkably consistent message. At the gross level, five of the six variants generated positive alpha in the range of roughly +2.2% to +3.5% annualized. That matters. It means there really is something there in the residual mean reversion mechanism.</p><p>But gross alpha is not what anyone gets to keep.</p><p>Once realistic trading costs are included, the picture changes sharply. The best net result came from the unhedged PCA specification, not because it had the strongest signal, but because it avoided the extra trading burden of rebalancing a hedge alongside every stock trade. The hedged PCA and ETF-based variants often had similar or even better gross signal content, but their fee bill was roughly double. That was enough to erase the edge.</p><p>This is the first major lesson of the whole exercise: in this framework, implementation dominates intuition. You do not get paid for having the prettier model. You get paid for surviving the fee drag required to express it.</p><p>That distinction is crucial because it changes the interpretation of the results. If gross alpha were absent, the framework would simply be broken. But gross alpha is present. What breaks is the route from signal to realized P&amp;L.</p><h3>The strategy’s strongest result is also its most misleading one</h3><p>The only baseline specification that finished above starting capital was the unhedged PCA variant. At first that appears to vindicate the simplest implementation: forget the hedge, keep the residual signal, trade the stock directly, and accept a little market exposure.</p><p>But looking inside the trade anatomy shows why that result is deceptive.</p><p>The long leg of the strategy earned about +5.96% annualized net. The short leg lost about -5.87%. Those numbers are so symmetrical that they almost cancel one another out mechanically, leaving behind the tiny +0.09% headline net result. This is not the signature of a clean market-neutral alpha stream. It is the signature of a strategy that is quietly leaning on conditional beta.</p><p>That matters because it means the best-performing variant is not really the cleanest expression of the signal. It is the variant in which the hedge has been removed and the market has been allowed back into the trade.</p><p>In other words, the strongest headline result is also the least pure.</p><p>The real signal is easier to see in the hedged variants, where the long and short legs are much more balanced and the per-leg win rates are actually quite respectable. But because those variants trade an extra hedge leg, they pay too much to monetize what they are seeing. So the irony is this: the cleaner implementations reveal the signal more honestly, and the dirtier implementation produces the best superficial performance.</p><h3>Where the framework still seems to work</h3><p>One of the more interesting findings in the report is that performance is not uniform across the cross-section.</p><p>The framework works relatively better in Energy, Technology, and Basic Materials. It works poorly in Real Estate, Consumer Cyclical, Communication Services, and several other sectors. That pattern is not random. It suggests that residual mean reversion works best where equity factor models can actually capture the dominant drivers of systematic risk, where idiosyncratic moves are more likely to be overreactions than structural repricings, and where the factor structure is stable enough to be estimated consistently through rolling windows.</p><p>That also helps explain why sectors like Real Estate and Financials behave poorly. Their key risk drivers are often not well summarized by equity-only factor constructions. Interest-rate sensitivity and regime shifts matter too much. In Healthcare, genuinely stock-specific events are often structural rather than transient, which is poison for a strategy that assumes reversion.</p><p>This is one of the places where the framework still earns respect. It is not universally dead. It appears to work under identifiable structural conditions. The problem is that those conditions are narrower than the textbook presentation encourages you to believe.</p><h3>Liquidity does not tell the story you expect</h3><p>A natural hypothesis is that the strategy should work better in less liquid stocks. Illiquid names ought to overreact more, face more impatient order flow, and therefore show stronger reversions once the pressure passes.</p><p>But that is not what the data show, at least not inside the S&amp;P 500 universe.</p><p>The liquidity breakdown produced no clear monotonic pattern. In the best-performing specification, the profitable buckets were the least liquid quintile and the second-most liquid quintile, while the middle bucket was the worst. Other variants produced different shapes entirely. The most defensible conclusion is that, within large-cap U.S. equities, liquidity is not the main discriminating variable for this strategy.</p><p>That does not mean liquidity is irrelevant in general. It means that testing for an illiquidity premium within the S&amp;P 500 is too narrow a laboratory. If there is a stronger liquidity story, it probably lives further down the capitalization spectrum.</p><h3>Regimes do not rescue the strategy</h3><p>Another tempting story is that the framework only works in the right market environment. Maybe it thrives when volatility is high. Maybe it needs more cross-sectional dispersion. Maybe correlation breakdowns create larger temporary dislocations.</p><p>There is some truth to pieces of that story, but not enough to save the strategy as implemented.</p><p>Cross-sectional dispersion does help. That makes intuitive sense: more dispersion means more names get pulled far enough away from their factor-implied levels to create tradable snap-backs. But the biggest regime effect in the entire analysis came from market breadth, and that is not good news. The strategy performed much better on days when most stocks were rising than on days when most stocks were falling. A genuinely market-neutral residual strategy should not care that much about whether the tape is broadly green or red.</p><p>That sensitivity is diagnostic, not redemptive. It shows that the portfolio is carrying conditional beta exposure, particularly in the decile-based long-short implementation. The regime cut therefore reinforces the main argument rather than softening it: the problem is not that the strategy happened to be tested in the wrong states of the world. The problem is that residual mean reversion is too weak relative to costs, and the baseline implementation leaks market exposure in the process.</p><h3>The most important finding: when the signal works, it works fast</h3><p>The cleanest and most valuable result in the entire study is about time.</p><p>Trades that closed within ten trading days were strongly profitable. Trades held between roughly eleven and twenty days were mediocre to flat. Trades that remained open beyond three weeks were destructive. This pattern was one of the most stable in the whole report. It appeared both in the single-stock trade anatomy and in the cross-sectional horizon sweep.</p><p>That tells you something fundamental about the signal. It is not a medium-term mean reversion effect. It is a short-horizon one, with a practical half-life of roughly one to two weeks.</p><p>This point changes everything.</p><p>A strategy can be conceptually correct and still be operationally wrong because it is holding the right trades for too long. That seems to be exactly what is happening here. The textbook Avellaneda-Lee implementation lets trades persist until the s-score has normalized or the backtest ends. In practice, that policy allows too many trades to drift out of the domain of temporary mispricing and into the domain of structural repricing. By then the model is no longer waiting for the market to correct a mistake. It is waiting for reality to reverse itself.</p><p>And reality usually does not cooperate.</p><p>If I had to reduce the whole report to one operational recommendation, it would be this: any deployable version of this framework needs a much harder time discipline than the textbook version implies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VrHY7o9yi4Ki5eyiEbmgnA.png" /></figure><p>The entry threshold is probably too permissive</p><p>The same logic appears in the entry signal.</p><p>The standard entry threshold of 1.25 generates the bulk of the trades, but those trades are too weak to clear transaction costs reliably. The best-performing bucket in the study was not the weakest signal region, and it was not the most extreme one either. The most attractive zone was the intermediate range, where the deviation was strong enough to matter but not so extreme that it was likely to represent a true structural revaluation.</p><p>That again fits the broader theme. The framework is real, but its tradable region is narrower than it looks. Too close to the threshold and the signal is too weak. Too far from equilibrium and the move may not be a mispricing at all.</p><p>The sweet spot exists, but it is not generous.</p><h3>Fees are not a detail. They are the whole game</h3><p>This is the uncomfortable part.</p><p>At zero fees, all six variants become profitable. At institutional-style fees around 2 basis points per side, most of them become meaningfully positive. At 10 basis points per side, which is a reasonable all-in assumption for many non-institutional implementations, the edge is mostly gone.</p><p>That does not mean the strategy is bad. It means the strategy is fee-sensitive to an unusual degree because it is built on frequent trading, small edges, and in many versions an extra hedge leg.</p><p>This is where a lot of quant research goes subtly wrong. People talk about alpha as though it were an intrinsic property of a signal, when in reality alpha is always the result of a signal passed through an execution regime. Change the execution regime and you may invert the ranking of the variants entirely. In my tests, the unhedged PCA variant won at high fee levels because it was cheaper to trade, but the hedged PCA variant overtook it once fees were lowered enough.</p><p>That is an important conceptual point. The question is not simply, “Does the signal work?” The question is, “Under what fee structure does the cleaner implementation of the signal become superior?”</p><p>For this strategy, that boundary matters a great deal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j4JEZANFhiM3ssYo9q-MvA.png" /></figure><p>The factor attribution does not offer an escape hatch</p><p>One possible interpretation of all this is that the positive or negative results are really just disguised exposure to standard risk premia. Maybe the residual strategy is secretly a market bet, or a size tilt, or a value tilt.</p><p>The Fama-French three-factor attribution does not support that view.</p><p>Across the variants, the market, SMB, and HML loadings were small and statistically insignificant. Factor contributions to returns were tiny, nowhere near large enough to explain the realized performance of the strategies. Even the best baseline variant, which showed slight positive raw net performance, became essentially zero after factor adjustment rather than strongly positive or strongly negative.</p><p>That is actually a useful result. It tells us that the framework is not simply rediscovering a standard risk premium. Nor is it blowing up because of some obvious hidden factor bet.</p><p>What it is doing is much more specific: it is harvesting a real but fragile mean reversion effect inside the idiosyncratic component, then struggling to keep enough of that effect after costs and imperfect implementation.</p><h3>The cross-sectional version says the same thing more cleanly</h3><p>I also tested a cross-sectional formulation of the strategy. Instead of treating each stock independently and trading whenever its own s-score crossed a threshold, I ranked the whole universe by s-score on each rebalance date, bought the bottom decile, shorted the top decile, and swept across different rebalance horizons. The advantage of this formulation is that it makes the turnover-versus-decay tradeoff much easier to see.</p><p>And the result was consistent with everything else.</p><p>At the gross level, the signal was real. The best in-sample gross Sharpe was respectable. At the net level, realistic costs consumed most of the advantage, and the best net outcomes were still too small to be compelling after model risk and sampling uncertainty. The key practical takeaway remained the same: if the framework is to be traded at all, it must live in a fee regime materially below 10 basis points per side and in a holding horizon of roughly ten to twenty trading days. Outside that joint envelope, the signal either decays or becomes uneconomic.</p><p>That convergence matters. It means the main conclusion does not depend on one particular implementation detail. Single-name and cross-sectional versions point to the same operational truth.</p><h3>So what survives?</h3><p>Quite a lot, actually.</p><p>What survives is not the fantasy version of statistical arbitrage, where a clever residual model automatically mints alpha once it is coded. That version does not survive contact with realistic execution. But a more disciplined version does.</p><p>The Avellaneda and Lee framework still appears to contain a genuine short-horizon residual mean reversion signal. It is strongest when the factor structure is estimable, when the move is truly temporary rather than structural, when the holding period is kept tight, and when costs are low enough not to consume the spread. The framework is therefore not dead. It is simply much narrower, more conditional, and more operationally demanding than its elegance suggests.</p><p>That is a less glamorous conclusion than many quants would like. But it is also more valuable.</p><p>Because there is a big difference between saying, “This classic strategy no longer works,” and saying, “This classic strategy still contains a real signal, but only inside a very unforgiving implementation envelope.”</p><p>The first conclusion is final. The second is actionable.</p><h3>Final thought</h3><p>My final view after working through the evidence is this: Avellaneda and Lee should still be studied, but not copied.</p><p>It remains one of the best pedagogical frameworks in statistical arbitrage because it formalizes a very deep trading intuition: price deviations only become tradable when you define the benchmark they are deviating from. That idea is alive and well.</p><p>What should not be copied uncritically is the textbook execution logic. In modern large-cap equities, residual mean reversion is too thin, too fast, and too fee-sensitive to survive a naïve implementation. Hedge too much, pay too much, or hold too long, and the edge disappears. Tighten the holding window, lower the fee burden, and be far more selective about where the signal is allowed to operate, and something more interesting begins to emerge.</p><p>That, to me, is the real lesson.</p><p>The framework was not wrong.</p><p>It was just much more fragile than it first appeared.</p><p>(The full report, is posted on my linkedin profile: <a href="https://www.linkedin.com/posts/oren-tapiero_statarb-analysis-activity-7452290568023629824-uvnh?utm_source=share&amp;utm_medium=member_desktop&amp;rcm=ACoAAACkQcABaHubpSDf5SH56ztvL3SEzcoaUbI">https://www.linkedin.com/posts/oren-tapiero_statarb-analysis-activity-7452290568023629824-uvnh?utm_source=share&amp;utm_medium=member_desktop&amp;rcm=ACoAAACkQcABaHubpSDf5SH56ztvL3SEzcoaUbI</a> )</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b6dac83b5dfa" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>