{"id":201178,"date":"2024-01-17T11:03:53","date_gmt":"2024-01-17T16:03:53","guid":{"rendered":"https:\/\/ibkrcampus.com\/?p=201178"},"modified":"2024-01-17T11:03:26","modified_gmt":"2024-01-17T16:03:26","slug":"peer-reviewed-theory-and-expected-stock-returns","status":"publish","type":"post","link":"https:\/\/www.interactivebrokers.com\/campus\/ibkr-quant-news\/peer-reviewed-theory-and-expected-stock-returns\/","title":{"rendered":"Peer-Reviewed Theory and Expected Stock Returns"},"content":{"rendered":"\n

The article “Peer-Reviewed Theory and Expected Stock Returns” first appeared on Alpha Architect<\/a> blog.<\/em>
<\/p>\n\n\n\n

As professor John Cochrane observed, the literature on investment factors now fills a veritable \u201cfactor zoo<\/a>,\u201d with hundreds of options. How do investors select from among this huge array of possibilities? In order to minimize the risk that outcomes result from data mining, in our book \u201cYour Complete Guide to Factor-based Investing<\/a>,\u201d Andrew Berkin and I established six criteria for a factor to be considered for investing. In addition to having provided a statistically significant premium, the premium must have been persistent across time and economic regimes; pervasive across asset classes, countries, regions, and sectors; robust to various definitions; have risk- or behavioral-based explanations for why the premium should be expected to persist in the future; and are implementable (survive transactions costs). While we preferred that the explanation be risk-based (because risk cannot be arbitraged away), we accepted behavioral-based explanations as long as there were identifiable limits to arbitrage that prevented sophisticated investors from correcting mispricings. All six criteria had to be met. <\/p>\n\n\n\n

By combining text analysis of publications with out-of-sample tests,\u00a0Andrew Chen<\/a>,\u00a0Alejandro Lopez-Lira<\/a>, and\u00a0Tom Zimmermann<\/a>, authors of the July 2023 study \u201cPeer-Reviewed Theory Does Not Help Predict the Cross-section of Stock Returns<\/a>,\u201d examined whether financial theory helped predict the cross-section of returns. They read the papers corresponding to 191 published predictors (from the 2022 study \u201cPublication Bias in Asset Pricing Research<\/a>\u201d by Chen and Zimmermann) and assigned each predictor to \u201crisk,\u201d \u201cmispricing,\u201d or \u201cagnostic\u201d based on the arguments made in the texts. The authors explained:<\/p>\n\n\n\n

\n

\u201cOf the 191 predictors we examine, 67% were published in The Journal of Finance, Review of Financial Studies, Journal of Financial Economics, Journal of Financial and Quantitative Analysis, Review of Finance, or Management Science. 22% were published in top accounting journals (The Accounting Review, the Journal of Accounting and Economics, the Review of Accounting Studies, and the Journal of Accounting Research). The remaining 11% were published in a wide variety of economics, finance, and accounting journals, including the Journal of Political Economy, Quarterly Journal of Economics, and Review of Economic Dynamics.\u201d\u00a0<\/em><\/p>\n<\/blockquote>\n\n\n\n

They began by noting:<\/p>\n\n\n\n

\n

\u201cIf risk-based theory helps us understand expected returns, then risk-based predictability should persist out-of-sample. At the very least, predictability that peer review attributes to risk should be more persistent than predictability derived from pure data mining.\u201d They added: \u201cEconomic theory should restrict this set in a way that tilts toward true predictability and therefore limits the impact of selection bias. Peer review further restricts the set of theories, as only the best theories should make it into the top finance journals.\u201d <\/p>\n<\/blockquote>\n\n\n\n

To address the data mining question, they compared the out-of-sample returns of risk-based theory to those from naive data mining by 29,000 trading strategies formed by sorting stocks on simple functions of 242 accounting variables. They explained:<\/p>\n\n\n\n

\n

\u201cThese strategies were formed by (1) dividing one variable by another or (2) taking first differences and then dividing, where the denominators were restricted to be variables that are positive for more than 25% of firms in 1963. Despite the complete lack of economics in this data mining exercise, we find this procedure generates substantial \u2018out-of-sample\u2019 returns. Sorting strategies into quintiles based on past returns each year and then trading the bottom quintile earns 50 bps per month, implying thousands of strategies with meaningful out-of-sample returns.\u201d<\/p>\n<\/blockquote>\n\n\n\n

Following is a summary of their findings: <\/p>\n\n\n\n