How New Information Technology Affects Asset Pricing Anomalies

Economic theory posits that the degree to which a signal is incorporated into asset prices is inversely related to the cost of acquiring that signal. Thus, a reduction in the cost of acquiring information—even information that is commonly referred to as “public”—should cause that information to be more fully incorporated into price.

The Efficient Markets Hypothesis, as originally formalized by Eugene Fama, drew a sharp line between “private” information (known only to some) and “public” information (known to all at no cost). But that line has always been blurry. Even information that is technically public—like a 10-K filing—can be costly to acquire or to process. Before 1993, for example, an investor who wanted a company’s accounting filings had to either travel to one of three SEC reference rooms (in Chicago, New York, or Washington, D.C.) or pay a commercial data vendor a substantial fee. The authors cite a 1992 SEC petition noting that one such vendor charged roughly $125 per month, plus connection and search charges that could run into the tens of dollars per query.

This matters because if acquiring or processing a signal is costly, standard theory predicts that only some investors will bother to do it. That means the signal won’t be fully reflected in prices—creating return predictability we call an “anomaly.”

David Hirshleifer and Liang Ma, authors of the January 2025 study “The Effect of New Information Technologies on Asset Pricing Anomalies,” investigated the causal effect on anomaly mispricing of improved access to financial disclosure signals using two natural experiments.

What the Authors Examined

The authors use two staggered technology rollouts as natural experiments:

EDGAR (1993–1996): The SEC’s phased introduction of electronic filing, which made corporate 10-Ks freely and instantly accessible online rather than requiring a physical visit or a paid subscription. This is a pure “access” shock—it didn’t change how hard the information was to interpret, just how easy it was to obtain.
XBRL (2009–2011): The SEC’s mandatory adoption of a standardized, machine-readable tagging format for financial statements. This is a pure “processing” shock—by this point, filings were already available on EDGAR, but XBRL made the data machine-readable, dramatically cutting the cost of extracting and analyzing specific line items at scale.

Because both technologies were rolled out in phases—EDGAR across 10 waves based largely on firm size, XBRL across three waves based on filer status—the authors can compare “treated” firms (those that just gained the technology) against “control” firms (similar-sized firms that hadn’t yet been treated) using a stacked difference-in-differences (DiD) design. This approach, following recent econometric research by Goodman-Bacon, avoids a known bias in traditional staggered DiD estimates that arises when already-treated firms are inadvertently used as controls.

Their tests were based on the most relevant subset of the 11 anomalies that are the focus of Stambaugh, Yu, and Yuan’s 2012 paper “The Short of It: Investor Sentiment and Anomalies,” as they are often attributed to mispricing. Among the remaining eight anomalies in the final set, five use annual accounting information that was hard to get before EDGAR and hard to process before XBRL: Accruals, Net Operating Assets, Investment to Assets, Asset Growth, and Gross Profitability. The other three are non-accounting anomalies (Momentum, Net Stock Issuance, and Composite Equity Issuance) that use market information (such as stock returns, number of shares outstanding, and market cap) that was already cheaply available through sources like Barron’s even before these technologies existed.

The logic of including the non-accounting group is important—it serves as a placebo test. If EDGAR and XBRL are really working through the channel the authors propose (specifically reducing the cost of accounting information), then anomalies that never depended on hard-to-get accounting data shouldn’t be affected.

Key Findings

EDGAR weakened accounting anomalies. The effect of EDGAR implementation on aggregate anomaly returns is statistically negative (about -0.4% per month for both raw and Fama-French 3-factor adjusted returns) for accounting anomalies—the return spread associated with these anomalies shrank meaningfully once a firm’s filings became available on EDGAR relative to firms that weren’t yet on the system.
EDGAR did not meaningfully affect non-accounting anomalies. The comparable coefficient for momentum, net stock issuance, and composite equity issuance was a small and statistically insignificant (about 0.1% per month)—exactly what you’d expect if EDGAR’s effect runs through easing access to accounting data rather than some broader market-wide shift.
XBRL weakened accounting anomalies by a similar magnitude to the EDGAR effect: -0.45% per month and −0.36% per month for raw and Fama-French three-factor adjusted returns, respectively (both statistically significant) despite the two technologies working through different channels (access versus processing cost).
The effect of XBRL adoption on aggregate anomaly returns is small in magnitude and statistically insignificant for non-accounting anomalies—supporting the idea that XBRL’s effect was concentrated in the accounting-based signals it was designed to make machine-readable.
The economic magnitudes of the effects of EDGAR and XBRL on accounting anomalies are similar, suggesting both that advances to modern information technology have improved market efficiency, and that easing the processing of information can be just as important as easing its availability. It’s also worth noting that the magnitude of the EDGAR and XBRL effects on accounting anomalies (roughly 0.36%–0.45% per month) is comparable to the average pre-treatment strength of those same anomalies (0.25%–0.37% per month). In other words, these technology shocks eliminated a very large share of the mispricing associated with treated firms relative to controls.

In a test of robustness, the authors extended their analysis to a broader set of 32 accounting anomalies and 34 non-accounting anomalies that are categorized as mispricing by peer review based on texts in the original papers (Chen, Lopez-Lira, and Zimmermann’s 2024 study “Does Peer-Reviewed Research Help Predict Stock Returns?”). The results are similar, supporting the prediction that new information technologies can alleviate mispricing and diminish anomalies.

Their findings led Hirshleifer and Ma to conclude: “We find that both EDGAR and XBRL reduce mispricing for accounting-based anomalies but not for non-accounting-based anomalies. The economic magnitudes of the effects on accounting-based anomalies are similar for EDGAR and XBRL. These results suggest that both easier access to and less costly processing of public information enhance market efficiency.” They added: “Our findings are consistent with limited investor cognitive processing power affecting market prices, and show that new financial information technologies help alleviate this constraint and make markets more efficient.”

Implications for Market Efficiency

Hirshleifer and Ma add nuance to our thinking about “public” information. Fama’s classic dichotomy treated information as either public (fully priced) or private (partially priced), but this study shows that “public” information exists on a spectrum of practical accessibility, and price efficiency moves along with that spectrum. Notably, the fact that XBRL’s effect (a pure processing-cost reduction) was just as large as EDGAR’s (a pure access-cost reduction) suggests that cognitive and computational processing costs are just as important a friction as raw availability. Simply posting a document online isn’t enough for markets to become fully efficient with respect to it—investors also need practical tools to extract and use the information at scale.

There’s also a broader interpretive question the authors address directly: if institutional investors could always afford data vendors or manual data collection, why weren’t they already arbitraging these anomalies away before EDGAR and XBRL arrived? The authors argue that trading costs and other limits to arbitrage—capital constraints, short-sale constraints, noise-trader risk—likely prevented full arbitrage even among sophisticated investors, and that broader market participation (including smaller investors who genuinely couldn’t afford $125-a-month data subscriptions) matters for how fully information gets incorporated into price.

Key Takeaways for Investors

Technology continues to be a powerful, ongoing force for market efficiency. EDGAR and XBRL were meaningful upgrades in their era, but the same logic applies to more recent developments—cheap computing power, natural language processing of filings, and now large language models capable of parsing footnotes and disclosures at scale. Each of these reduces the cost of extracting signal from public filings, and this paper’s findings suggest each such advance should be expected to further erode accounting-based anomalies over time.
Momentum and other market-data-based anomalies aren’t necessarily subject to the same erosion. Because non-accounting anomalies didn’t respond to either EDGAR or XBRL, investors shouldn’t assume that all anomalies decay at the same rate or for the same reasons. The information underlying momentum, for example, was already cheap and easy to get from sources like stock tickers and financial newspapers—so improvements in accounting-data technology had nothing to erode there. That said, other research (e.g., the post-publication decay literature) suggests trading-cost reductions and increased attention from the investing public can still erode these signals over time, just via a different mechanism than the one studied here.
Costs of information still matter—for some investors more than others. The persistence of anomalies before EDGAR and XBRL, despite the theoretical ability of institutional investors to pay for the same data, is a reminder that limits to arbitrage (trading costs, capital constraints, short-sale constraints) can allow mispricing to persist even when sophisticated capital is technically available to correct it. This is a useful caution against assuming that “smart money” alone is sufficient to guarantee efficient prices; broad-based information accessibility and processing improvements appear to matter as an independent channel.
The case for skepticism toward complex, accounting-signal-dependent trading strategies grows stronger. For advisors and investors evaluating strategies pitched as exploiting accounting-based anomalies, this paper is another reminder to ask how much of any historical premium has already been competed away by the same technological forces that made the underlying data more accessible in the first place.

Finally, in our book “The Incredible Shrinking Alpha“ Andrew Berkin and I showed that there were four main themes that were making it increasingly difficult for active managers to outperform appropriate risk-adjusted benchmarks:

Academic research has been converting what was once alpha into a systematic factor (beta).
The pool of victims that can be exploited has been shrinking (due to the increasing market share of passive investing).
The competition has been getting tougher.
The supply of dollars chasing alpha has increased.

Hirshleifer and Ma provide a fifth theme: increasing information availability and ease of information processing has made markets more efficient, reducing or even eliminating pricing anomalies, thus increasing the hurdles to successful active management.