<?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 Atmajitsinh Gohil on Medium]]></title>
        <description><![CDATA[Stories by Atmajitsinh Gohil on Medium]]></description>
        <link>https://medium.com/@atmajitgohil?source=rss-55712a22e7df------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*IxTonRJsjwJSvAUi.</url>
            <title>Stories by Atmajitsinh Gohil on Medium</title>
            <link>https://medium.com/@atmajitgohil?source=rss-55712a22e7df------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 18:55:07 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@atmajitgohil/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[Nowcasting using Internet Search data]]></title>
            <link>https://medium.com/@atmajitgohil/nowcasting-using-internet-search-data-20589c2a983f?source=rss-55712a22e7df------2</link>
            <guid isPermaLink="false">https://medium.com/p/20589c2a983f</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[forecasting]]></category>
            <category><![CDATA[nowcasting]]></category>
            <category><![CDATA[linear-regression]]></category>
            <dc:creator><![CDATA[Atmajitsinh Gohil]]></dc:creator>
            <pubDate>Mon, 01 Apr 2024 05:51:11 GMT</pubDate>
            <atom:updated>2024-04-01T05:51:11.988Z</atom:updated>
            <content:encoded><![CDATA[<p>A summary of nowcasting has been provided in the prior blog article. The primary objective of Nowcasting is to predict the current economic environment using the data available at time <em>t</em>. In the present blog post we explore the literature on use of internet search data to perform Nowcasting.</p><p>The present blog explores the use of internet search trend to predict the mortgage refinancing¹. The Mortgage bankers Association publishes a weekly index on the refinancing activity which has a delay of one week. Given the high penetration of internet in USA it is possible that individuals intending to perform refinancing would perform a search on the internet prior to refinancing their mortgage. It is possible that many individuals who search the internet may not actually refinance — this is a limitation of the analysis that is not adequately addressed by the authors.</p><p>The following image provides a comparison between the search count of the term “mortgage refinance” and the actual mortgage refinance index. The internet search closely tracks the mortgage refinance application index.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/491/1*G_wK5M1KEvS_mGE8b-tpiQ.png" /></figure><h3>Data Collection:</h3><p>The authors collect the internet search data using the following:</p><ol><li>the search term “mortgage refinance” for the geographic area the United States and in the subcategory “home financing,”</li><li>A lag of the applications index and</li><li>The ten-year Treasury yield and its one-period lag</li></ol><h3>Model:</h3><p>The following table provides a simple regression of the refinance index as the dependent variable and the search term , refinance index lag 1 and 10 Year treasury yield and its lag as the independent variables. The R — Square reported is 34%.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/546/1*7hvo5ftWgsxEkPMA4txcAw.png" /></figure><h3>Conclusion</h3><p>Nowcasting using internet search would be a handy tool for economists for predicting refinancing activity. The refinance activity prediction would aid the federal reserve bank and regulators in accurate policy making. Though the study highlights the limitations of using internet search the blog post introduces a new source of data in the Nowcasting literature.</p><h3>Reference:</h3><ol><li><a href="https://libertystreeteconomics.newyorkfed.org/2012/01/forecasting-with-internet-search-data/">https://libertystreeteconomics.newyorkfed.org/2012/01/forecasting-with-internet-search-data/</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=20589c2a983f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Nowcasting Economic data]]></title>
            <link>https://medium.com/@atmajitgohil/nowcasting-economic-data-a194c83e45ce?source=rss-55712a22e7df------2</link>
            <guid isPermaLink="false">https://medium.com/p/a194c83e45ce</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[random-forest]]></category>
            <category><![CDATA[nowcasting]]></category>
            <category><![CDATA[artificail-intelligence]]></category>
            <category><![CDATA[forecasting]]></category>
            <dc:creator><![CDATA[Atmajitsinh Gohil]]></dc:creator>
            <pubDate>Thu, 28 Mar 2024 13:10:18 GMT</pubDate>
            <atom:updated>2024-03-28T13:10:18.925Z</atom:updated>
            <content:encoded><![CDATA[<h3>Nowcasting:</h3><p>Understanding the current economic environment is important for policy makers and government agencies. The effectiveness of policies such as controlling inflation rate and interest rate is based on availability of current economic data. However, the economic data such as GDP and inflation is released with a lag which does not provide a an accurate depiction of the current economic environment. This critical issue is resolved using Nowcasting which is a methodology to predict the current economic activity based on the data released today.</p><h3>Applications of Nowcasting:</h3><p>Nowcasting has been successfully implemented by many Governmental agencies and Central banks. The Federal Reserve Bank of Atlanta¹ publishes GDPNOW. The GDPNOW provides an estimate of GDP growth rate for United States of America. The Federal Reserve of Cleveland² provides Inflation Nowcasting. Similar Nowcasting approaches have been adopted by many developing and developed economies.</p><p>Nowcasting framework has also been utilized for non-economic purposes. The UNHCR³ (UN Refugee Agency) is using Nowcasting methodology for Refugee and Asylum Seekers.</p><h3>Methodology:</h3><p>The advances in faster computing power and machine learning techniques have successfully led in development and implementation of Nowcasting. The following image provides a 3 step framework for nowcasting:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/784/1*ja6VMfGkkSc3hXMCCb0F7A.png" /><figcaption>Source: NOWCASTING WORLD TRADE WITH MACHINE LEARNING: A THREE-STEP APPROACH⁴</figcaption></figure><p>The authors in the paper⁴ provide a 3-step framework that includes Pre-Selection, Factor Extraction and Non-Linear regression. The authors compare the performance of various machine learning models for nowcasting world trade data and provide a combination that produces the lowest error and improves accuracy. Within the 3-step framework the authors have identified a combination of LARS (Pre-Selection), PCA (Factor Extraction) and Macroeconomic Random Forest (Non-linear Regression) perform better than other popular models.</p><h3>Nowcasting Model Performance:</h3><p>The authors compare the model accuracy of various Machine Learning models using Root Mean Square Error (RMSE). A lower RMSE implies a higher model accuracy and vice versa. As shown in the image below the RMSE of the Macroeconomic Random Forest methodology is lowest for current period (<em>t</em>), backtesting (<em>t-1, t-2</em>) and forecasting (<em>t+1</em>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/681/1*w-FR5FW_UmH9HgSAgqvQqw.png" /><figcaption>Model Accuracy</figcaption></figure><h3>Conclusion:</h3><p>The use of nowcasting for predicting current economic environment has assisted both central banks and government agencies in framing policies that are appropriate and based on current data and no lag data. The tremendous growth in computing power and development of Machine Learning and Artificial Intelligence (AI) techniques have paved way to development of nowcasting methodologies that were not accessible before.</p><h3>References:</h3><ol><li><a href="https://www.atlantafed.org/cqer/research/gdpnow">https://www.atlantafed.org/cqer/research/gdpnow</a></li><li><a href="https://www.clevelandfed.org/indicators-and-data/inflation-nowcasting">https://www.clevelandfed.org/indicators-and-data/inflation-nowcasting</a></li><li><a href="https://www.unhcr.org/refugee-statistics/insights/explainers/nowcasting-refugees-asylum-seekers.html">https://www.unhcr.org/refugee-statistics/insights/explainers/nowcasting-refugees-asylum-seekers.html</a></li><li><a href="https://www.nber.org/system/files/working_papers/w31419/w31419.pdf">https://www.nber.org/system/files/working_papers/w31419/w31419.pdf</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a194c83e45ce" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Governance through Computing Power]]></title>
            <link>https://medium.com/@atmajitgohil/ai-governance-through-computing-power-e500ad84be25?source=rss-55712a22e7df------2</link>
            <guid isPermaLink="false">https://medium.com/p/e500ad84be25</guid>
            <category><![CDATA[risk-management]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[computing-power]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai-governance]]></category>
            <dc:creator><![CDATA[Atmajitsinh Gohil]]></dc:creator>
            <pubDate>Mon, 25 Mar 2024 04:08:15 GMT</pubDate>
            <atom:updated>2024-03-25T04:08:15.967Z</atom:updated>
            <content:encoded><![CDATA[<p>Can we use computing power to effectively govern AI? Can we frame policies around computing power that would allow us to implement an effective AI policies? These key questions are answered by the authors in their research post article <em>Computing Power and the Governance of AI¹. </em>Before diving into the the topic of using computing power as a tool for AI governance it is critical to understand the concept of computing power from a non-technical perspective.</p><h4>FLOPS as a measure of Computing Power:</h4><p>Floating Point Operations Per Second or FLOPS in short is a quantifiable metric that is used to measure the computational efficiency and capacity of an AI system. The FLOPS denotes the number of floating point operations executed per second. This metric essentially measures the computational prowess and processing speed of AI-driven algorithms and models, thereby serving as an indicative benchmark for assessing their performance².</p><p>Flops on a computer with exactly 1 CPU can be calculated as follows³:</p><p>FLOPS = (Cores) * (Cycles / Seconds) * (FLOPs / Cycles)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/257/1*wS1u27UlnlI9SuMyiLwNMA.png" /><figcaption>Source: <a href="https://en.wikipedia.org/wiki/FLOPS">Wikipedia</a></figcaption></figure><p>The computing power has increased substantially since 2010. According to the Epoch database Gemini Ultra, released in 2023, has the highest computing power of approximately 10²⁷ FLOPS or Ronna FLOPS.</p><h4>Characteristics of Computing Power:</h4><p>The authors have outlined four characteristics such as detectability, excludability, quantifiability and supply chain of computing power that can be leveraged to perform AI governance.</p><ol><li>Detectability: AI development is resource intensive which requires large capital investment in the form of data centers, computer chips and consumption of power. Hence it is easy to detect AI development.</li><li>Excludability: AI development requires hardware&#39;s which can be restricted.</li><li>Quantifiability: As highlighted in the above section it is possible to quantify the computing power which allows for easy reporting and verifiability.</li><li>Concentrated Supply Chain: AI chip production requires multiple steps such as chip design, manufacturing of fabrication equipment, and chip fabrication. Hence, very few players in the market have the ability to produce these chips causing an inelastic supply.</li></ol><h4>AI Risk Management :</h4><p>The four characteristics of computing power can be leveraged to govern AI through visibility, allocation and enforcement.</p><ol><li>Visibility: The characteristics such as detectability and quantifiability allows for easy visibility. An increased visibility allows for easy detection of players in AI development. The governance policies such as reporting of computing power and registry of chips would allow the government to monitor development of AI. The government can also implement noninvasive techniques of using publicly available information to monitor AI development.</li><li>Allocation: The computing power characteristics such as excludability and quantifiability can be used by government agencies and policymakers to prioritize allocation of resources for AI development. The policies aimed at resource allocation and increased subsidy to sectors such as education, health, climate change and agricultural efficiency will advance AI development in these sectors. As computing power is quantifiable government agencies can procure AI resources through alliances that can be used for AI development in beneficial sectors and avoid misuse.</li><li>Enforcement: The enforcement of laws to restrict the misuse of AI can be used to regulate and develop AI. The supply chain characteristics can be used to restrict Chip development through export restrictions.</li></ol><h4>Conclusion:</h4><p>The computing power can provide effective policy tools that can be used for AI governance. It should be noted that as AI algorithms and technology matures the computing power tool for AI governance becomes less effective. Lastly, by limiting computing power the development of AI for harmless application is also curtailed. Hence, use of computing power as a policymaking tool should be carefully reviewed to ensure its effective implementation.</p><h4>Reference:</h4><ol><li>Computing Power and the Governance of AI: <a href="https://www.governance.ai/post/computing-power-and-the-governance-of-ai">https://www.governance.ai/post/computing-power-and-the-governance-of-ai</a></li><li><a href="https://www.larksuite.com/en_us/topics/ai-glossary/flops">https://www.larksuite.com/en_us/topics/ai-glossary/flops</a></li><li><a href="https://en.wikipedia.org/wiki/FLOPS">https://en.wikipedia.org/wiki/FLOPS</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e500ad84be25" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>