Paul Ma

We estimate a dynamic model of voluntary disclosure, using annual management forecasts of earnings, that features a manager with price motives and an uncertain but persistent information endowment. Our estimates imply that: (i) managers face disclosure frictions 35% of the time; (ii) conditional on being informed, managers withhold information 17% of the time; and (iii) conditional on being silent, managers possess information 24% of the time. Managers’ strategic withholding motives increase investors’ uncertainty about earnings by 3%. We find that managers’ price motives reduce strategic withholding by one-third, in response to investors’ increased skepticism in the event of non-disclosure.

Applying a "co-search" algorithm to Internet traffic at the SEC's EDGAR website, we develop a novel method for identifying economically-related peer firms. Our results show that firms appearing in chronologically adjacent searches by the same individual (Search Based Peers or SBPs) are fundamentally similar on multiple dimensions. In direct tests, SBPs dominate GICS6 industry peers in explaining cross-sectional variations in base firms' out-of-sample: (a) stock returns, (b) valuation multiples, (c) growth rates, (d) R&D expenditures, (e) leverage, and (f) profitability ratios. We show that SBPs are not constrained by standard industry classification, and is more dynamic, pliable, and concentrated. Our results highlight the potential of the collective wisdom of investors - extracted from co-search patterns - in addressing long-standing benchmarking problems in finance.

Relative total shareholder returns (rTSR) has become the predominant metric to isolate managers' idiosyncratic performance. Among firms that explicitly use rTSR in relative performance contracts, 60%―those that choose specific peers as benchmarks―select rTSR metrics that do a remarkable job of filtering out the systematic component of returns in adherence to the informativeness principle. However, firms that choose index-based benchmarks retain substantial systematic noise in their rTSR metrics. We document that the selection of noisy benchmarks is associated with compensation consultants' preferences, which are uncorrelated with observable firm attributes. Firms with weak governance are more likely to choose indexes, not because of opportunism, but because they do not adequately scrutinize outside experts' advice. Collectively, our findings provide a new explanation for why some executives are evaluated based on systematic noise, and novel evidence on how compensation consultants can impact firms.

The vast majority of China’s fertility decline predates the famous One Child Policy – and instead occurred under its predecessor, the Later, Longer, Fewer (LLF) fertility control policy. In this paper, we first study LLF’s contribution to marriage and fertility behavior, finding that the policy reduced China’s total fertility rate by about 0.9 births per woman, explaining only 28% of China’s modern fertility decline. Given son preference, we then consider the parallel issue of sex selection, which also emerged prior to the One Child Policy (when prenatal selection was not technologically feasible). We find that LLF increased the use of male-biased fertility stopping rules from 3.25% to 6.3% of couples – and that it contributed to the early emergence of postnatal neglect of girls in modern China, rising from none to 0.3% of births (implying 210,000 previously unrecognized missing girls). Considering Chinese population policy to be extreme in global experience, our results demonstrate the limits of population policy’s ability to reduce fertility – and its potential for unintended consequences.

We address whether mandatory forecasts of earnings announcement dates are informative and what are the informational tradeoffs between mandatory and voluntary forecasts. We find China mandatory forecasts predict actual earnings announcement dates and yet-to-be-announced firm performance, and the market reacts to the initial and revised forecasts accordingly. Regarding informational tradeoffs we find the following. China mandatory forecasts are informative, even by firms less likely to issue a voluntary forecast; this information is unavailable in a voluntary regime. The act of US voluntary forecasting and its timing reveal information incremental to the forecasted announcement date, which is unavailable in a mandatory regime. Perhaps surprisingly, US voluntary and China mandatory initial forecasts convey a similar amount of earnings news, which is noteworthy because the China forecasts are issued substantially earlier and suggests the amount of information in the act and timing of voluntary forecasts is small.

We apply modern machine learning techniques to characterize disclosure misclassification by public companies. We find that 12-26% of disclosures are misclassified; those concerning material definitive agreements, executive or director turnover, and delistings are most commonly misclassified. Using EDGAR search traffic data, we provide evidence that misclassification is associated with less investor attention. Through this attention channel, misclassification leads to a significant and persistent impact on absolute market returns. For misclassified filings, search traffic is 4-12% lower and absolute market reactions are 46-79 bps smaller. Consistent with strategic motives, misclassification is more likely for negative news and when market attention is high.

In knowledge-based economies, many businesses enterprises defy traditional industry boundaries. In this study, we evaluate six "big data" approaches to peer firm identifications and show that some, but not all, "wisdom-of-crowd" techniques perform exceptionally well. We propose an analytical framework for understanding when crowds can be expected to provide wisdom and show, theoretically and empirically, that their efficacy is related to crowd sophistication and task complexity. Consistent with this framework, we find that a "crowd-of-crowds" approach, which combines EDGAR user co-searches and analyst co-coverage, dominates other state-of-the-art methods for identifying investment benchmarks.

This paper examines whether investors correctly distinguish qualitative information from promotional language in press releases related to material events of US public firms. For a variety of material events, firms are required to issue a Form 8-K, but 37% of the time also voluntarily issue a press release concerning the same event, half of which occur prior to the 8-K filing date. Using textual analysis, I find that firms are more likely to issue a press release if the underlying 8-K tone is positive, and that tonal differences between the 8-K and the press release are driven in part by quotes from officers. I also find economically significant responses in firms' stock returns to tonal language in the 8-K, as well as to tonal differences between the two disclosures. To verify whether my strategy of comparing the press release against the 8-K is isolating the effects of promotional language or additional information, I test and find evidence of an initial positive reaction but subsequent negative drift from positively toned press releases. This implies that investors may have initially responded to both information and spin. Nominating investor inattention as a possible mechanism for overreaction, I use novel search traffic micro-data from the SEC EDGAR website and detect lower 8-K search intensity in the presence of a press release. Together, my results are consistent with some investors overestimating the degree of substitutability between the two disclosures and thus failing to readjust expectations accordingly.

We identify the presence of high frequency arbitrageurs in the US treasury market through intraday exchange outages. Evidence complementing our identification shows that order cancellation behavior also changed during the outage, consistent with arbitrageurs' profit maximization motives. Our estimates suggest that arbitrageurs represent approximately 69 to 94% of the quote depth in the spot treasury market. In addition, their presence seems to have large effects for the bid-ask spread of the 30-year treasury bond, which is the most illiquid product within its class.