Hurtado-Sanchez (1978) set out to test the technical analyst's traditional view of high short interest as a bullish indicator, but his results apply to the academic models as well. He wondered if the inclusion of hedging and arbitrage-motivated trades in short-interest data obscures the information content of speculative short sales. Rather than test directly for the prevalence of these trades, he considers whether short interest predicts future returns using a sample of stocks from the Standard & Poor's 425 Industrials of 1966 and 1967. He fails to detect any evidence that levels or changes in absolute short interest, the SIR, or RSI, can predict future performance in individual stocks. He does find, however, that stocks with high (low) return performance experience increases (decreases) in short interest in the following month. His conclusion is that short-interest data contain no information about future returns, but short sales help stabilize the market by adding to selling pressure after prices have risen.

Figlewski (1981) was one of the first to consider the implications of Miller's overpricing hypothesis. Figlewski assumes that observed levels of short interest proxy for the amount of unfavorable information excluded from market prices as a result of the constraints on short sales. In other words, a relatively high level of short interest in a stock indicates that short interest would have been even higher yet, if unconstrained. He also refines Miller's overpricing hypothesis by pointing out that rational investors, with knowledge of the effects of short-sales constraints, would not overprice some stocks without underpricing others. Thus, he hypothesizes that high (low) levels of short interest predict overpricing (underpricing) in individual stocks. Figlewski's appeal to rational expectations is somewhat of a precursor to Diamond and Verrecchia except Figlewski allows for informational inefficiency at the firm level. That is, in his model investors have yet to learn that short interest proxies for the amount of unfavorable information excluded from market prices. Diamond and Verrecchia get around the assumption of firm-level inefficiency by focusing on the information content of unexpected changes in short interest.

He finds mixed support for this hypothesis in a sample of Standard & Poor's 500 Index stocks from the years 1973 to 1979. Specifically, a short position in the stocks ranking the highest on RSI outperforms a long position in the lowest RSI stocks by a statistically significant amount, but only if the short seller captures the interest on the proceeds from the short sale. Of course, most small traders receive no interest on short-sale proceeds, and even large traders must pay a loan free as compensation to the lender. Excluding the interest on proceeds, the mean return to stocks ranking highest in terms of RSI is actually positive in the post-ranking month, although insignificant.

The inability of both Hurtado-Sanchez and Figlewski to detect compelling evidence of return predictability from short interest suggests that hedging and arbitrage-motivated trades may be obscuring any information content in the data. Examples of such trades include the arbitrage of going long convertibles or warrants and short the converting common stock, the arbitrage of mergers (that is, going long the targets stock while shorting the acquirer's stock), and general "pairs trading." Pairs trading is a general term used to describe strategies that involve buying a stock that is thought to be underpriced, for any of a number of reasons, and shorting a statistically paired stock to neutralize risk and possibly to further enhance return. The need to understand the motivations of short sellers took on added importance with the introduction of Diamond and Verrecchia's previously mentioned work. They, of course, indicate that large unexpected increases in short interest predict negative future returns because short sellers are better informed. They also claim that the information content in short interest is obscured for stocks that have traded options.

Brent, Morse, and Stice (1990) considered the motivations of short sellers using random samples of 200 NYSE stocks from the years 1981 to 1984. Their tests confirm the results of Hurtado-Sanchez in that changes in RSI fail to predict future returns, but stocks with high returns subsequently experience large increases in RSI. The latter finding is in direct opposition to one of the key assumptions behind the technical analysts' bullish view of short interest: that short selling supposedly increases in down markets. Thus, it appears that short sellers are attempting to anticipate mean reversion in returns. They also observe that stocks with high short interest tend to have high betas, traded options, and listed convertible securities. They therefore conclude that hedging and arbitrage, as opposed to speculation, motivates a material amount of short selling.

Another hedging strategy that may obscure information in short interest is "shorting against the box" (that is, selling short a stock already held long) at the end of the year to delay capital gains to the following year. Using NYSE and Nasdaq short interest data from 1995 to 1999, Arnold, Butler, Crack, and Zhang demonstrate the popularity of this strategy prior to the Tax Payer Relief Act of 1997. The Act disallowed this practice as a means to delaying taxes, and they find that year-end short interest declined significantly with the introduction of the Act. They also show that the Act had the effect of strengthening the negative relation between changes in a stock's RSI and its return in the following month. This clearly indicates that short interest announcements contain information about subsequent returns, in the manner of Diamond and Verrecchia, as long as information-motivated trades make up an adequate proportion of the short interest.

Senchack and Starks (1993) tested the predictive power of short interest with an event study on a sample 2,419 stocks selected so as to be less susceptible to the problem of obscured information content. They begin with all NYSE and AMEX stocks, whose short interest was published in the Wall Street Journal from 1980 through 1986. The sample is then purged of stocks reported to be the subject of arbitrage activities, although this does not account for pairs trading and shorting against the box. They also eliminate all observations in which the reported increase in short interest, from the previous month, is less than 100%. This is done to better reflect the model of Diamond and Verrecchia, which applies only to large, unexpected increases in short interest. Finally, the sample differentiates between stocks that have traded options and those that do not.

Senchack and Starks point out that buying puts and writing calls is a low cost alternative to short selling, and this means that any unfavorable private information about a stock may be observable from option premiums and volumes, well before the short interest announcement. Note that the short interest figures may be relatively unaffected if put writers hedge by selling short. They find that stocks without traded options have a small but statistically significant negative price reaction to the announcement of large percent increases in short interest. The cumulative negative returns over both five- and nine-day event windows are slightly less than one-half of 1%. In addition, the larger the percent increase in short interest, the more negative is the price reaction to the announcement. Stocks with traded options, on the other hand, display no significant reaction to announcements of large percent increases in short interest. These results support both Diamond and Verrecchia's prediction that large, unexpected increases in short interest are bearish signals, as well as the claim that traded options obscure the information content in short interest announcements.

Figlewski and Webb (1993) take a somewhat different approach in their study of the effect of options on short-sale constraints. They recognize that options decrease the costs of effectively going short and suspect that this improves informational efficiency by making constraints on short sales irrelevant. Note that the combination of reduced trading costs and increased informational efficiency should weaken, if not eliminate, the ability of short interest to predict future returns.

Using samples of Standard & Poor's 500 stocks from the 1970s and 1980s, they establish that the options market is actively used as a complement to short selling. Stocks with traded options have significantly higher RSI levels than stocks without traded options, and the introduction of traded options in a stock tends to increase the stock's RSI. They also find that option premiums tend to be higher in puts than in calls for stocks with high levels of RSI. These results suggest that option trading enables more negative information to enter the market, and impact stock prices, than would have otherwise. The impact on stock prices occurs as a result of put writers selling short to hedge, as well as from the arbitrage when the puts become expensive relative to the calls. This arbitrage involves writing the put, buying the call, and shorting the stock.

For stocks with high levels of RSI, Figlewski and Webb find that those without traded options earn negative returns, in the month after the announcement, but these negative returns are not significantly less than the returns to the stocks with traded options. Senchack and Starks, of course, find this difference to be significant, as is expected if options actually improve informational efficiency. The discrepancy is likely due to the cleaner sample used by Senchack and Stark, as well as the concentrated focus of their five- and nine-day event windows. In addition, Senchack and Starks analyze only large percent changes in short interest, while Figlewski and Webb analyze levels of RSI.

A study by Choie, Huang, and James (1994) supports the view that large percent changes in short interest signal more about short-term returns than do high levels or large increases in short interest. They find that a short position in the stocks with the largest percent increases in short interest, as reported by the Wall Street Journal in the years 1988 to 1991, earned a mean return of more than 1% in excess of the S&P 500 Index in the month following publication. This is about double the excess return from shorting the stocks with the highest short interest levels or the largest SIRs. In addition, the stocks with the largest simple increases in short interest actually outperformed the S&P 500 Index, on average, in the month following publication. This suggests that percent changes are more difficult to predict and therefore are unexpected in the manner of Diamond and Verrecchia.

Most of the work we have reviewed, up until now, finds that large changes and, to a lesser extent, high levels of short interest predict small negative returns in the month or days after the announcement. However, these returns are statistically significant in only a few cases, and their economic significance is even less certain. Probably the most compelling evidence comes from Senchack and Starks, who focus on large percent increases in short interest and find support for the predictions of Diamond and Verrecchia.

Focusing on the short-term price reaction to large percent increases in short interest is an appropriate test of Diamond and Verrecchia, but it is not clear that any of the above work provides a fair test of Miller's overpricing hypothesis because it results from short-sale constraints. Thus, it will not be eliminated by a short-interest announcement, whether the focus is on short interest levels or changes. The price adjustment process may be much slower, and therefore, detectable only over longer horizons. This implies that short interest may need to accumulate for some unspecified time before any correction occurs.

Asquith and Meulbroek (1995) investigate the long-term returns to NYSE and AMEX stocks with very high RSI at some point from 1976 to 1993. While the previously mentioned work relies on short interest data reported in the financial press, Asquith and Meulbroek construct their own comprehensive data set. This is done because the financial press reports this data only for stocks with high levels or large changes in short interest. In August 1995, for example, the Wall Street Journal reported short interest only for those stocks with positions larger than 300,000 shares or changes of more than 50,000 shares from the previous month. Asquith and Meulbroek, on the other hand, wish to analyze RSI, not large levels or changes in short interest. This is because RSI reflects the supply of shares outstanding in the denominator, and they believe that supply together with demand (the numerator in RSI) will dictate the longer-term return. (Note that relying on the Wall Street Journal might preclude some stocks with high RSI if they do not also satisfy the reporting cutoffs.)

Asquith and Meulbroek focus on the excess returns to stocks that attain relatively high levels of RSI for as long as the high levels persist and for up to two years afterwards. In this way, they avoid the timing problem of earlier studies that requires precise alignment of the price reaction with the short interest announcement. They also point out several reasons why traded options may not obscure the information content in short interest. First, interviews with practitioners, including hedge fund managers, reveal that establishing large short positions with put options on hard-to-borrow stocks is more expensive and offers less liquidity than direct short selling. In addition, although one may be forced to cover a short sale early, there is no definite expiration date as with options, and this can be a serious disadvantage when speculating on a possible downturn in a stock. Finally, very few stocks under heavy selling pressure have listed put options. For stocks with RSI at or above the 95th percentile, less than 2% have listed put options traded.

Slightly under 24% of the stocks in the sample reach the 95th percentile of RSI at some point from 1976 to 1993; the RSI at this percentile is roughly 2.5%, on average, over the period. Stocks that attain this 95th percentile, or above, earned mean size-adjusted returns of −18% while remaining at or above this level, plus an additional −23% in the two years subsequent to falling below this level. The excess returns to stocks at the 99th percentile of RSI are even more stunning, but only about 7.5% of the stocks ever reached this level, and it is probably safe to assume that it is almost impossible to borrow these stocks. Note also that these returns do not include the rebate interest that institutional short sellers may receive. The statistically significant negative excess returns persist over the entire 18-year period, and they are even more negative for firms that are heavily shorted for more than one month.

Although it may be difficult to borrow stocks with RSI at or above the 95th percentiie, these returns would still appear to be of economic significance. Even if these stocks cannot be sold short, a high RSI should still serve as a sell signal to those who are long the stock, and at a minimum, these results would seem to relegate to myth status the traditional technical analysts' view that high short interest is a bullish indicator. In addition, the slow reaction of stock prices, that takes months if not years, is strong support for the overpricing hypothesized by Miller, as well as Figlewski.

Desai, Ramesh, Thiagarajan, and Balachandran (2002) extend the work of Asquith and Meulbroek to Nasdaq market stocks with comprehensive monthly short interest data obtained directly from the Nasdaq for the years 1988 to 1994. Based on improved methods from the performance measurement literature, they measure long-term excess returns by controlling for market-to-book ratios and momentum, as well as size and beta. Their results suggest that short sellers target highly liquid stocks whose prices have recently improved relative to fundamentals.

Stocks with RSI of 2.5% or more earn mean excess returns of −6.6% within one year and −8.8% within two years of attaining this level. Upon falling back below this 2.5% level, they continue to earn negative excess returns, on average, of −7.3% within one year and −11.2% within two years. These negative returns increase with higher RSI levels. They also find that the heavily shorted stocks are liquidated or forced to delist with a higher frequency than their size, book-to-market, and momentum-matched control firms. Farinella, Graham, and McDonald (2001) verify these results independently. Thus, Asquith and Meul-broek's conclusion that high short interest signals bearish sentiment about future returns applies to the Nasdaq market as well as the NYSE and AMEX.

Although these studies detect highly negative long-term returns without removing the stocks with traded options, it would be a mistake to assume that traded options have little or no effect on overpricing. Danielson and Sorescu (2001) of options introductions between 1981 and 1995 clearly shows that options improve informational efficiency by reducing the cost of short selling. They find that prices decline and short interest increases for stocks just after their options are first listed. The increase in short interest appears to be due to the purchase of puts by previously constrained short sellers whose intent is then transferred into short sales by the hedging activities of the put writers. As long as the marginal put writer is a market professional, with transactions cost advantages at short selling, the put contracts will represent a reduction in the cost of constructing an effective short position.

Diamond and Verrecchia predict that the lower costs of options will obscure the information contentofshort interest, but Danielson and Sorescu's price declines are unique to the overpricing hypothesized by Figlewski and Miller. Also consistent with the overpricing hypothesis, Daniel-son and Sorescu find that the price declines are larger in stocks with higher betas and greater dispersion of investor opinions, as proxied for by volume, return volatility, and analysts' forecasts. They suggest, however, that these predictable price declines are not exploitable because of the high cost of short selling these stocks prior to the listing of their options.

The magnitude of these negative returns, reported by Asquith and Meulbroek as well as Desai, Ramesh, Thia-garajan, and Balachandran, raises an important question. That is, beyond the point that high short interest predicts negative future returns, what factors determine the level of short interest in a stock? The fact that excess returns remain negative for up to two years suggests that accumulated short selling does, eventually, move prices in the direction of fundamentals. Understanding the determinants of short interest may offer some insights into identifying short-sale candidates early, before short interest increases until costs are prohibitive or borrowing becomes impossible. Of course, acting early is less costly, but there is also the added risk of acting too soon. The negative returns may take longer, or they may not materialize at all.

It is well known that stocks with relatively low fundamental-to-price ratios experience systematically lower returns in the future. Using data on NYSE and AMEX stocks from 1976 to 1993, Dechow, Hutton, and Meulbroek (2001) document that short sellers target stocks that rank low based on ratios of cash-flow-to-price, earnings-to-price, book-to-market, and value-to-market. A stock is considered "targeted" if its RSI is 0.5% or higher. Short positions in these stocks earn positive excess returns in the year after they are targeted, as prices fall, and the ratios mean-revert. Further more, short sellers refine this strategy in three ways by avoiding stocks (1) that are expensive to short, such as small stocks with low institutional ownership and high dividends, (2) with low book-to-market ratios that appear justifiable due to high growth potential, and (3) with justifiably low fundamentals. These motives are confirmed by a telephone survey of major global hedge fund managers whose responses indicate that they short sell to profit from overpriced stocks.

Gintschel investigated the determinants of short interest in all the Nasdaq stocks eligible for margin trading between 1995 and 1998. Proxies for the float (that is, the supply of shares available to borrow), such as market capitalization and turnover, explain almost 60% of the cross-sectional variation in RSI. The significant time-series determinants of short interest are firm size, turnover, put option volume, as well as variables relating to technical and fundamental strategies, including future operating performance. He finds that short interest is equally sensitive to both positive and negative innovations in value and operating performance, suggesting it is motivated by hedging, while the short interest attributable to past returns is motivated by overpricing.

From an expectations model based on these findings, Gintschel computes unexpected changes in RSI and finds a significantly negative mean return of about 0.5% in the 15 days after the announcement of unexpectedly high RSI. He also detected a negative mean return of about 1% from the time short interest data are collected until the actual announcement, which indicates considerable leakage. In addition, he suggests that the negative long-term returns reported by Asquith and Meulbroek and Desai, Ramnesh, Thiagarajan, and Balachandran may be due to very high market capitalizations and low book-to-market ratios, rather than overpricing.

Boehme, Danielson, and Sorescu argue that tests of overpricing should use a two-dimensional framework based on Miller (1977). Recall that Miller indicates that binding short-sale constraints and high dispersion of investor beliefs are both necessary conditions for overpricing. Using RSI as a proxy for short-sale constraints, and return variance as well as share turnover as proxies for dispersion of beliefs, Boehme, Danielson, and Sorescu find that controlling for both yields low returns in constrained, high-dispersion Nasdaq and NYSE stocks between 1988 and 1999. Specifically, these stocks have a mean raw return of zero and a mean excess return of −20% over a one-year horizon, although this underperformance is less severe in stocks with traded options. (Considering either short interest or dispersion of beliefs separately does not yield significant excess returns.) Boehme, Danielson, and Sorescu suspect, however, that much of this underperformance cannot be arbitraged due to the high costs of short selling and the difficulty in borrowing these shares.

Pownall and Simko (2003) examine the fundamentals of stocks that are targeted by short sellers in "short spikes" (that is, abnormally large increases in short interest), as announced in the Wall Street Journal during the years 1989 through 1998. They also consider the price response to the announcement of a spike in short interest as well as whether the short sellers are profitable. The stocks targeted by short sellers are not materially different, in terms of fundamentals, from the population of NYSE firms during the period immediately prior to the spike. However, in the year subsequent to the short spike, the targeted stocks experience significant declines in key earnings-based fundamentals, such as earnings-to-price and earnings growth.

Their sample-wide mean excess return over the five-day intervals beginning with the announcement of the short spike is negative but small. For individual stocks, excess returns are more negative the larger the price run-up in the months prior to the spike. The profitability of short selling is measured by computing excess returns from the date the spike is announced until short interest returns to normal levels. The mean return for stocks that revert to normal levels of short interest within four months is −1% and significant, with all of this return coming in the month the reversion occurs. The sample-wide mean cumulalive excess return is −#1% and significant; however, most of this profit is attributable to the one-third of the sample that takes more than nine months to revert to normal levels of short interest. (Over 75% of the sample stocks revert to normal levels within less than a year.)

These cumulative excess returns are significantly larger for stocks without traded options, for stocks with RSI greater than 2.5%, and for spikes that occur prior to 1994 (when hedge fund trading began in earnest). This last finding is of particular importance since the large post-announcement returns reported by Asquith and Muel-broek and Desai, Ramesh, Thiagarajan, and Balachandran were observed from samples that end in 1993 and 1994, respectively. The implication is that hedge fund managers are either exploiting (through speculation) or obscuring (through hedging) the information content of short interest such that it no longer persists for long periods, post announcement.

Pownall and Simko conclude that the profits to trading on short spikes are small, except in extended positions, which may be difficult to maintain and thus are more risky. This is similar to Boehme, Danielson, and Sorescu's conclusion, as well as that of Gintschel. Although it would appear that the emergence of hedge funds has eroded much of the highly negative pre-1994 returns, it may be slightly premature to dismiss the post-1994 returns as un-exploitable. Instead, it would be better to more carefully consider the various costs of short selling.

In an earlier section, we briefly described the constraints on short sales: (1) the direct monetary costs of borrowing shares, (2) the difficulty (or impossibility) of establishing a short position, (3) the risk that the short position cannot be maintained, and (4) the legal and institutional restrictions on short selling. Now we wish to more carefully consider items 1, 2, and 3 since these are costs that limit the arbitrage of information contained in short interest data. Legal and institutional restrictions, in item 4, constrain short selling, but they do not represent a cost that an individual short seller actually faces.

The direct monetary cost of short selling is reflected in the rebate rate the lender of the stock pays to the borrower. Recall that the borrower sells the stock and the lender then has the use of the short-sale proceeds. Thus, the rebate rate represents the stock lender's cost of accessing funds less a compensating loan fee for lending the stock. Although rebate rates are usually positive, they can be negative if a stock is in such high demand (to borrow) that the loan fee is greater than the cost of funds. Rebate rates apply almost exclusively to institutional investors. Individual investors usually receive no interest on the proceeds from their short sales.

There is no centralized market for lending shares in the United States, and rebate rates are not publicly available. However, the activities of a large institutional lending intermediary during 2000 and 2001 are revealed in a study by D'Avolio (2002). He finds that fewer than 10% of the stocks this institution had available to loan are so-called specials, which have loan fees above 1%. The value-weighted loan fee across the entire available supply of shares is 0.25%. The average loan fee for specials is 4.3%, but fewer than 10% of these specials (less than 1% of all available stocks) are in such high demand that their rebate rates are negative.

For the stocks in the highest decile of short interest, D'Avolio reports an average loan fee of just under 1.8%, while about 33% of these stocks are specials. Stocks in the second highest short interest decile have an average loan fee of 0.8% and about 15% of these stocks are specials. Unfortunately, we do not know if the specials with high short interest experienced lower future returns than the general population of high-short-interest stocks. We do know, however, from Jones and Lamont (2002) that stocks with low or negative rebate rates have high market-to-book ratios and low subsequent returns, consistent with overpricing. Their results are based on a centralized market for lending stocks that was operated on the floor of the NYSE from 1919 to 1933. When stocks were newly listed on this lending market, they were overpriced by more than can be explained by the direct monetary costs of short selling. Jones and Lamont suggest that some other constraint on short selling must be limiting the arbitrage of this apparent opportunity.

The most obvious candidate is difficulty in borrowing the shares. However, Geczy, Musto, and Reed (2002) find that at least some of the profits to a number of popular shorting strategies are available to a hypothetical small investor who cannot short specials nor receive rebate interest. Their data are from a major institutional equity lender for 1998 and 1999. Unfortunately, they do not consider strategies based on short interest. Study Chen, Hong, and Stein suggest that overpricing survives because most institutional investors are restricted from short selling, and the rest of the market simply cannot absorb the opportunities. If this is true, it may bode well for the exploitation of carefully constructed short interest strategies that consider the accumulation of short interest over time. However, D'Avolio points out that loan fees are sticky in these decentralized lending markets; if so, stocks under increasing demand may be rationed prior to becoming specials, and this too could explain the Geczy, Musto, and Reed's results.

If short sellers worry that the risks of an early recall are high, or about being caught in a short squeeze, then they will require a premium for risky arbitrage. D'Avolio reports that the unconditional probability of a recall is low, with only 2% of the stocks on loan recalled in a typical month of his sample, but he also notes that recalls often occur when lenders receive negative information about a stock, which causes them to recall the shares, either to sell them or to reprice the loan. The possibility that negative information, possibly in the form of a rumor, could result in a recall is potentially unnerving for a short seller, and this introduces noise trader risk as an additional limitation to risky arbitrage. That is, a lender may rationally recall shares based on how less than fully rational investors may react to news, rather than based on fundamentals. Some short sellers request the identity of a potential lender to minimize the possibility of such a recall.

It is clear that constraints on short selling result in overpricing. It is also apparent from the studies by Gintschel, Boehme, Danielson, and Sorescu, and Pownall and Sirnko that even the post-1994 short interest data contain some information about future returns. Although there is no direct evidence, it would appear from D'Avolio as well as Geczy, Musto, and Reed that the monetary costs of short selling are probably not large enough to render short interest data unexploitable, at least not totally. It may, however, be difficult to borrow shares with high short interest, and possibly even more difficult to maintain the short position for long enough to realize a profit. In addition, D'Avolio points out that there is considerable risk associated with the early recall of a short position. It follows that these results may be viewed as consistent with market efficiency, at least to the extent that arbitrage opportunities are pursued to the limits of the costs and risks.

It is worth emphasizing that the existence of overpricing does not necessarily imply that short interest data contain information. Persistent overpricing relies on Miller's claim that the high costs of short selling constrain the less optimistic investors from trading based on their information, so that the market clearing price is determined by the overly optimistic investors. High short interest is a proxy for high costs only to the extent that short interest would have been proportionally that much higher, if unconstrained. Clearly, some stocks have low short interest precisely because short selling them is relatively costly.

The other academic justification for analyzing short interest comes from Diamond and Verrecchia's rational expectations model, which relies on short sellers with superior information. In their model, overpricing occurs only when the current level of short selling is higher than anticipated, and the entire correction comes with the short interest announcement that follows. It follows from Diamond and Verrecchia that higher frequency reporting of short interest, or transparency in short-sales transactions, should improve the informational efficiency of the U.S. stock markets. Next, we consider whether improvements are likely to actually result from any such changes.

Aitken, Frino, McCorry, and Swan (1998) were the first to provide evidence of the information content in short-sales transactions. Their data are from the Australian Stock Exchange for the years 1994 to 1996. This exchange reports transactions-level data, including short-sales information, to brokers and institutions on-line in real time. They report that short sales cause a rapid reassessment of price, with a mean of −0.2% within 15 minutes or 20 trades. There is less of a reaction to short sales associated with hedging activities, just as Diamond and Verrecchia would predict.

Aitken, Frino, McCorry, and Swan interpret their results as evidence that transparent short sales convey information as suggested by Diamond and Verrecchia. Note that this is claiming more than just short sellers have superior information. This is claiming that the execution of a short sale in this transparent market must immediately be recognized as an informed trade by other market observers who then, in turn, quickly sell long (or possibly short), and the price then, moves accordingly. In other words, the price moves directly as a result of other traders reacting to the short, seller's perceived information, rather than as a result of the short seller's actual information. Of course, the short seller does have to be informed if market efficiency is to improve as a result of transparency.

Angel, Christophe, and Ferri (2002) use daily transactions data from late 2000 to show that short sellers in Nasdaq-listed stocks have the ability to predict the direction of future earnings surprises as well as stock returns. But does this mean that the U.S. stock markets should become more transparent and issue more frequent and detailed reports about short sales?

The problem is that the very price adjustment process that should make a transparent market more efficient, that of Diamond and Verrecchia, is also a process that is ripe for manipulation and abuse. For example, almost daily we hear of short sellers being accused of "ganging up" on some stock in the hopes up driving its price down and then exiting at the opportune moment. Imagine how much easier this type of manipulation would be in a market with transparent short sales. This might result in the marginal short seller being a noise trader rather than an informed trader. In which case, the market would be less efficient than before. Finally, greater transparency can only address temporary mispricing that is consistent with rational expectations, as in Diamond and Verrecchia's model. Greater transparency does not reduce the costly constraints on short selling that drive the persistent overpricing Miller's model predicts. Thus, transparency may be of little benefit given that there is considerable support for Miller's overpricing hypothesis.