# Information and insurer financial strength ratings: do short sellers anticipate ratings changes?

ABSTRACTRatings of financial institutions have been shown to provide informational value as stock prices generally decrease in response to ratings downgrades. Moreover, insurer's stock prices have been observed to decrease 2 days prior to downgrades, suggesting that informed trading occurs during the predowngrade period. This study examines the trading activity of short sellers surrounding insurer financial strength ratings. We show that short selling is abnormally high during the predowngrade period--indicating that short sellers can predict rating downgrades. Interestingly, we find that predowngrade short selling is driven by stocks of insurers with the most transparent balance sheets. This result suggests that while short sellers can predict rating downgrades generally, the opaqueness of an insurer's assets and liabilities can inhibit informed trading during the predowngrade period.

INTRODUCTION

Research regarding the insolvency risk of financial institutions has received considerable attention, with good reason due to recent insolvencies in financial institutions. Doherty and Phillips (2002) argue that insurers attempt to market their financial strength (IFS) ratings as a signal of the firm's financial strength. Pottier and Sommer (1999) suggest that investors use these ratings to measure the risk of insurers, while Parekh (2006) suggests that insurers with ratings above some specified threshold are more popular than other insurers. Halek and Eckles (2010) find that stock prices of insurance companies tend to move in the direction of ratings changes, particularly for unfavorable changes, thus indicating that ratings provide informational value to the market. Further, Halek and Eckles (2010) show that returns during the 2 days prior to downgrades are significantly negative, suggesting that the market begins to move as informed investors begin to trade before the announcement of the downgrade is made public.

The objective of this study is to examine a subset of arguably informed traders--short sellers--around IFS ratings announcements. Specifically, we test whether the price reaction precipitated by ratings downgrades, upgrades, and affirmed ratings are anticipated by short sellers. Prior work generally finds that short-selling activity contains information about future price movements as current short-selling activity relates inversely to subsequent stock returns (Diamond and Verrecchia, 1987; Senchack and Starks, 1993; Aitken et al., 1998; Desai et al., 2002; Boehmer, Jones, and Zhang, 2008; Diether, Lee, and Werner, 2009). However, observing a negative relation between short selling and subsequent returns is not tantamount to finding that short sellers are privately informed, per se. In fact, Engelberg, Reed, and Ringgenberg (2012) show that the negative relation between daily short selling and next-day returns is driven by the ability of short sellers to process information that is already public.

The level of private information contained in short selling has been recently debated. Some research indicates that short sellers are consistently able to predict negative announcements, such as earnings announcements and analyst recommendations (Christophe, Ferri, and Angel, 2004; Christophe, Ferri, and Hsieh, 2010). However, other studies have shown that short-selling activity is not abnormally high prior to negative news events as short sellers are generally reactive as opposed to being proactive prior to firm-specific announcements (Daske, Richardson, and Tuna, 2005; Blau and Pinegar, 2012; Blau and Wade, 2012; Engelberg, Reed, and Ringgenberg, 2012).

While our study is motivated by this debate in these foregoing studies, examining the trades by short sellers around IFS rating announcements is particularly appealing for two reasons. First, while insurers are often rated on an annual or quarterly basis, IFS ratings are not usually announced on a fixed calendar date and are therefore less predictable than other types of announcements such as earnings or analysts' recommendations. Second, ratings are focused on insurance companies that vary in their level of transparency as their asset and liability structure is focused in different lines of insurance business and differs in the level of uncertainty (Ross, 1989; Baranoff and Sager, 2002; Zhang, Cox, and Van Ness, 2009). The cross-sectional variation in transparency of insurer liabilities and assets may adversely affect the ability of short sellers to correctly predict upcoming ratings because of uncertainty in properly evaluating the insurance company. Thus, the insurance industry and specifically IFS rating announcements provide a robust framework for testing whether short sellers trade on private information or have a superior ability to process already-public information.

Using a sample of 165 A.M. Best ratings announcements between January 1,2005 and December 31, 2006, we test whether short selling is unusually high in the period directly prior to IFS rating announcements. Our tests show abnormally high short selling prior to downgrades, after controlling for other factors that might affect the level of short selling. In additional tests, we find abnormally low short selling in the days prior to the ratings upgrades and relatively normal short selling in the days prior to announcements when ratings are affirmed (no change). Combined with our results that show short selling is unusually high during the period prior to downgrades, these results again suggest that short sellers have a unique ability to successfully predict the announcement of IFS ratings changes. Our findings also contribute to the results in Halek and Eckles (2010) by providing new evidence of informed trading in the days prior to ratings changes.

At first glance, these findings suggest that short sellers may be trading on private information before IFS ratings are announced. If nonpublic information is observed from the ratings agency, then the abnormal predowngrade short selling may be possibly considered insider trading. To disentangle whether short sellers are trading on private information or short sellers are unusually sophisticated in predicting downgrades, we condition our tests on the opaqueness of the insurer's assets and liabilities. To the extent that short sellers are trading on information leaked from the ratings agency, the abnormal predowngrade short selling is expected to be independent of the insurers' balance sheet opacity. However, if sophisticated short sellers are able to correctly identify insurers with decreasing financial strength, according to financial statement information, then the predowngrade short selling is likely to be driven by insurers that are most transparent.

Results show that the abnormal predowngrade short selling is driven primarily by stocks of insurers with the most transparency in both assets and liabilities. This finding suggests that the most transparent insurers attract the most short-selling activity prior to IFS rating downgrades. This result makes the case for insider trading by privately informed short sellers less compelling. Further, this finding provides an important contribution to literature as sophisticated investors are constrained in their trading by uncertainty caused by balance sheet transparency. When examining the abnormally low preupgrade short selling, we do not find a significant relation between preupgrade short selling and balance sheet opacity. Nor do we find that the relatively normal levels of short selling prior to announced nonchanges are affected by balance sheet opacity.

The rest of this article follows with the next section reviewing related prior literature. The following section develops our hypotheses, and sample and data are introduced in the subsequent section. The next two sections present our empirical methodology and discuss our results. The final section concludes.

PRIOR LITERATURE

Short Selling

Diamond and Verrecchia (1987) initially conjecture that short sellers are sophisticated investors and possess information about future firm performance and the true value of stocks. This assertion has been empirically supported by numerous studies that find short selling is inversely related to subsequent returns (Senchack and Starks, 1993; Aitken et ak, 1998; Desai et al., 2002; Boehmer, Jones, and Zhang, 2008; Engelberg, Reed, and Ringgenberg, 2012; Christophe, Ferri, and Angel, 2004; Christophe, Ferri, and Hsieh, 2010). However, observing a negative relation between short selling and subsequent returns is not necessarily equivalent to finding short sellers to be privately informed.

Short sellers' ability to predict negative returns may arise from their superior ability to process public information. Indeed, Engelberg, Reed, and Ringgenberg (2012) show that the return predictability of short sellers is markedly higher on days with information-rich announcements than on nonevent days, mainly driven by short sellers' superior ability to process public available information. Therefore, determining whether short sellers are privately informed requires examining short selling behavior before an information-rich event. Christophe, Ferri, and Angel (2004) and Christophe, Ferri, and Hsieh (2010) examine shorting activity prior to earnings announcements and analyst recommendations, respectively, and find that short selling is abnormally high prior to both unfavorable earnings announcements and downward analyst recommendation changes. When examining short selling around insider sales, Khan and Lu (2008) find abnormal short selling prior to insider sales. These previous findings substantiate the assertion that short sellers are sophisticated traders around information-rich events.

While there is a foundation of research suggesting short sellers are sophisticated, a recent stream of literature suggests that short sellers are no more sophisticated prior to informational events than other traders (Blau and Pinegar, 2012; Blau and Wade, 2012; Boehmer and Wu, 2013; Chakrabarty and Shkilko, 2013). Boehmer and Wu (2013) show that short sellers are not able to predict negative announcements and instead increase their shorting activity in response to announcements. Blau and Wade (2012) find the short-selling patterns surrounding both analyst downgrades and upgrades are remarkably symmetric, indicating that short sellers during the prerecommendation period are not unusually informed about the direction of upcoming recommendation changes. Their findings indicate that short selling prior to analyst recommendations is more likely speculative than informed. Blau and Pinegar (2012) find that short selling surges after both positive and negative announcements and that short selling immediately before negative announcements is less able to predict future returns than short selling during more normal times. Chakrabarty and Shkilko (2013) only find abnormal short selling on days with insider sales and the event-day short selling is not able to identify the insider sales that have the largest future stocks price decline, suggesting that the ability of short sellers to predict the negative news in insider trades is selective at best. These studies begin to question the informativeness of short sellers prior to negative news events. (1)

IFS Ratings

IFS ratings are the summary measures of insolvency risk (Pottier and Sommer, 1999). The rating provides a rating agency's opinion of the insurer's overall financial strength and ability to meet its policyholder obligations. IFS ratings have been related to a myriad of characteristics, such as capitalization, liquidity, and size (Pottier, 1998).

IFS ratings are assigned to both individual companies and to consolidated groups of insurance firms. These ratings are important because they influence the price insurers can charge for their policies (Doherty and Phillips, 2002).

Insurance company ratings are vitally important to consumers, insurers, investors, regulators, and insurance brokers and agents (Pottier and Sommer, 1999; Doherty and Phillips, 2002; Parekh, 2006). Insurance consumers use IFS ratings in determining from which insurance companies they purchase coverage and/or determining the cost they are willing to pay for insurance from their chosen company. The ratings provide value to the insurers who use the IFS ratings for advertising purposes in order to convey the company's financial strength and ability to meet obligations to their policyholders. Often during the individual insurance purchasing process, brokers and agents recommend coverage based on the ratings provided for a specific company, whereas corporate insurance consumers require that all their insurers be highly rated.

While the IFS ratings are utilized for different purposes, the IFS ratings contain new information that may be of interest to individuals or companies incorporating the ratings into their decision-making process. The informational content of ratings is apparent when looking at the reaction the capital markets have to a ratings change. Halek and Eckles (2010) hypothesize that a rating agency possesses superior information relative to the public and that its ratings announcements add to the public information related to an insurer. In testing their hypotheses, Halek and Eckles find that stock prices of insurance companies tend to move in the direction of ratings changes, particularly for unfavorable ratings changes, thus indicating that ratings provide informational value to the market.

With regard to the information contained in a ratings change, investors may be concerned with rating changes due to the potential changes in insurers' future cash flows. Doherty and Phillips (2002) suggest that during the period in which A.M. Best changed its ratings standards, insurers significantly increased their working capital. Furthermore, Doherty and Phillips argue that losing a high IFS rating had a significant impact on an insurer, and that rating agencies play an important part in reducing the asymmetric information between the insurers and consumers. Cummins and Danzon (1997) observe that insurance premiums are positively related to IFS ratings, while Pottier (1998) indicates that adverse rating changes had significant predictive power for forecasting life insurer insolvency. These previous findings combined with those of Halek and Eckles (2010) indicate that ratings provide information to investors on the financial strength of an insurer.

Insurer Opaqueness

The opaqueness of financial institutions, such as insurance companies, banks, and investment funds, has been the topic of research examining information-related factors and information asymmetries between financial institutions and investors. Ross compares the opaqueness of banks, insurance companies, and mutual funds and suggests that banks and insurers contain more asymmetric information in their asset composition than mutual funds. Additionally, Ross contends that insurance companies and banks are among the most opaque because managers have informational advantages about firm operations and, specifically, the level of risk in the firm's asset structure. Flannery, Kwan, and Nimalendran (2004) examine the relative opaqueness of various assets in bank portfolios and show that asset opaqueness affected the adverse selection costs. Overall, the literature suggests that insurance companies and banks present the greatest degree of information asymmetry between claimholders and the financial institution with regard to the institution's assets (Polonchek and Miller, 1999).

However, there are differences in opaqueness between banks and insurance companies. While both banks and insurance companies have a relatively similar asset opaqueness structure, their liability structures differ significantly. Banks' liabilities are normally well classified with regard to the monetary sum and length of exposure. However, insurance companies are unique in this manner, since their liabilities are far less certain due to the unpredictability of the length of the claims payout and the final overall payout. Accordingly, insurers' liabilities are associated with a much larger degree of information asymmetry than banks. Moreover, uncertainty about the length and amount of claims payouts differs between lines of business, which affects the degree of information asymmetry between policyholders, investors, and the insurer.

Prior research has identified lines of business that tend to be more opaque for both property-casualty (P/C) and life-health (L/H) insurers. Babbel and Merrill (2005) suggest the intricate nature and opaqueness of insurance policies allow managers for both P/C and L/H insurers to generate ambiguous financial measures of liabilities, surplus, and reserves. Phillips, Cummins, and Allen (1998) separate P/C lines of business into long-tailed and short-tailed lines, depending on the length of the claim payouts. Their results show the price of insurance is inversely related to the riskiness of the firm, and that these results are stronger for long-tail lines of business than for short-tail lines. Colquitt, Hoyt, and McCullough (2006) indicate P/C insurers increase the information asymmetry by utilizing greater discretion in setting loss reserves.

In examining L/H insurers, Baranoff and Sager (2002) suggest that accident and health lines contain more asymmetric information than annuities because of the uncertainty of when claims will be paid out. Baranoff and Sager also suggest that group lines of business contain more asymmetric information than individual lines. Zhang, Cox, and Van Ness (2009) separate lines of business into opaque and transparent lines. The authors then test the effects of opaqueness on the adverse selection component of the bid-ask spread. Their findings suggest that the opaqueness of a firm's liabilities directly affects the adverse selection component of the spread. This observation indicates that there is greater information asymmetry in insurers with more opaque assets and liabilities. Because of this asymmetric information caused by insurer opaqueness, sophisticated investors may have difficulty predicting rating changes.

HYPOTHESES DEVELOPMENT

Halek and Eckles (2010) document that change in IFS ratings contain important information that begins to impact stock prices prior to the announcement of the downgrade. Additionally, Pottier and Sommer (1999) argue that ratings are predictable based on public information; we expect that short sellers, who are shown to be informed in the literature, will anticipate IFS rating downgrades and increase shorting activity during the predowngrade period. However, Engelberg, Reed, and Ringgenberg (2012) argue that the information contained in short-selling activity is attributable to short sellers' ability to process already-available public information instead of their ability to predict the information. Under the assertion of Engelberg, Reed, and Ringgenberg, short selling may be more reactive than proactive around downgrades.

We therefore distinguish between two hypotheses. First, if short sellers can anticipate downgrades, then we expect to observe abnormally high short-selling activity prior to the downgrade. Further, if short sellers are able to identify the information before the information becomes publicly available, then the abnormally high short selling during the predowngrade period should be orthogonal to the opaqueness of insurers' assets and liabilities. We denote our first hypothesis as the informational advantage hypothesis.

Informational Advantage Hypothesis: Predowngrade short selling is abnormally high and is independent of insurer opacity.

It is possible that short sellers can predict the upcoming IFS ratings change without obtaining private information. For instance, short sellers may be able identify factors that determine ratings (Pottier and Sommer, 1999). However, the identification of these factors is likely more difficult for insurance companies with highly opaque asset and liability portfolios. Therefore, finding abnormally high predowngrade short selling in the least opaque insurers is consistent with what we denote as the sophisticated trading hypothesis.

Sophisticated Trading Hypothesis: Pre-downgrade short selling is abnormally high and is driven by insurers with the least opaque asset and liability portfolios.

SAMPLE AND DATA

We obtain short-sale data in response to the Securities and Exchange Commission's Regulation SHO from January 2005 to December 2006 and aggregate the short-sale data to the daily level. (2) From the Center for Research in Security Prices (CRSP), we obtain daily volume, prices, returns, shares outstanding, and market capitalization. Following Diether, Lee, and Werner (2009), we restrict our sample to stocks that trade every day of the time period (January 2005 to December 2006), have price greater than $2, and have a CRSP share code of 10 or 11. We obtain the lines of business, liability, and asset data from the NAIC database and follow Zhang, Cox, and Van Ness (2009) and Baranoff and Sager (2003) in defining opaque and nonopaque liabilities and assets in P/C and L/H lines of business, respectively.

A.M. Best's Key Rating Guides and A.M. Best's Insurance Reports provide insurer ratings data for our sample period. In defining our ratings sample, we follow the methodology of Halek and Eckles (2010). Insurers in our sample must be rated twice during the 2005-2006 sample period. Based on the nature of the insurance industry, frequently multiple individual insurance companies may be held by a single publicly traded company. In these instances, the group rating for an insurer is utilized; while for companies that are not a member (or a singular member) of a group, the rating for that single company is used. Furthermore, if there is not a group rating for a publicly traded insurer, but companies within the group have the same ratings, this rating is used.

The sample consists of 165 publicly traded insurers (89-PC, 76-LH) with a total of 25, 14, 126 downgrades, upgrades, and affirmed A.M. Best ratings, respectively. While we would prefer to analyze more events, we are restricted in extending our time period because of the limitations of the Regulation SHO data. It should be noted, over our 2-year time period the distribution of ratings events is consistent with previous annual ratings events over a 10-year period (Halek and Eckles, 2010). In addition, we focus our analysis on A.M. Best ratings, as A.M. Best ratings provide stronger results when reporting cumulative abnormal returns around announcements when compared to those of S&P or Moody's (Halek and Eckles, 2010; Eckles and Halek, 2011).

We use two (standardized) measures of short selling as our dependent variables--short ratio (SR) and short turnover (STO). Following Diether, Lee, and Werner (2009) we define SR as the daily number of shares sold short for a stock divided by the total number of shares traded in the stock during the same day. We calculate STO as the daily short volume scaled by the number of shares outstanding. There are variations in the literature as to which measure of short selling should be used in empirical testing. Diether, Lee, and Werner suggest the short ratio measure is much less skewed than the other measures of short-selling activity. Additionally, Christophe, Ferri, and Hsieh (2010) suggest that distributional differences in short-selling activity around announcement events may be due to unusually high or low trading volume prior to the announcement, and that measuring short selling as a percentage of trading volume would result in a relatively constant measure. Recent event studies by Chakrabarty and Shkilko (2013) and Christophe, Ferri, and Hsieh (2010) break from the short ratio methodology and use an alternative method to scale short volume. Chakrabarty and Shkilko define short volume as an abnormal ratio of short to nonshort volume, whereas Christophe, Ferri, and Hsieh scale short volume by the number of shares outstanding. A similar measure is used in prior work (Senchack and Starks, 1993; Dechow et al., 2001; Desai et al., 2002; Asquith, Pathak, and Ritter, 2005; Christophe, Ferri, and Hsieh, 2010).

Our primary variables of interest are the IFSR and the insurer's opaqueness in their lines of business (i.e., assets or liabilities). We define these indicator variables as Rating, Lopaque, and Aopaque. The model specification for the ratings announcement is either an action of a downgraded (Down), upgraded (Up), or affirmed (Affirm) rating. The Rating variable equals one if day t is the day of a ratings action (i.e., downgrade, upgrade, or affirm), and zero otherwise.

We follow Zhang, Cox, and Van Ness (2009) and Baranoff and Sager (2003) in defining opaque and nonopaque lines of business and assets for PC and LH lines of business, respectively. P/C opaque is the ratio of premiums written in opaque PC lines of business relative to total premiums written in all PC lines. L/H opaque is the percentage of premiums written in opaque LH lines of business relative to premiums written in all LH lines. We combine these measures and arrive at a liabilities measure Lopaque defined as the percentage of premiums written in all opaque lines of business relative to total premiums. By definition Lopaque lines are: aircraft, automobile, fidelity, medical malpractice, surety, workers compensation, accident and health (only), and other liabilities. We follow Hannery, Kwan, and Nimalendran (2004) and Zhang, Cox, and Van Ness (2009) in our definition of asset opaqueness (Aopaque). Consistent with these prior studies we define relatively opaque assets as mortgage loans, real estate investment, private placement loans and bonds, premium notes, premiums receivable, other investments, reinsurance recoverable on loss and loss adjustment expense payments, and reinsurance ceded.

The control variables in our analysis are widely included in the microstructure literature. Diether, Lee, and Werner (2009) and Boehmer, Jones, and Zhang (2008) control for Size as market capitalization influences the short selling of stocks. The authors suggest smaller stocks may have less of a following by analysts and therefore may experience more trading by informed investors than larger stocks. Additionally, Arnold et al. (2005) use Size as a proxy for institutional holdings, suggesting that institutions, which are generally lenders of shares to short sellers, are likely to hold larger stocks. Therefore, a security's market capitalization will affect the level of short selling. However, if asymmetric information exists more in small-cap stocks, then short selling may be negatively related to size. Consistent with this argument, Diether, Lee, and Werner find that short sellers are more informed in smaller stocks.

Diether, Lee, and Werner (2009), show that daily short volume is positively related to contemporaneous returns, indicating that short sellers are contrarian traders in past returns. Contemporaneous returns represent the movement of the stock price during the 3 days prior to the current trading date. Additionally, Diether, Lee, and Werner suggest short-sale volume has positive serial correlation and short selling is related to turnover. We therefore control for contemporaneous returns ([ref.sub.i, t-3, t-1]) and lagged short selling and trading activity ([SSRi.sub.i t-8, t-4] and [SSTO.sub.i, t-8, t-4]) on the right-hand side of Equations (3)-(6).

Previous findings document that price volatility (return volatility) positively (negatively) affects the level of short selling (Diether, Lee, and Werner, 2009; Lamoureux and Lastrapes, 1990). Lamoureux and Lastrapes (1990) show that return volatility approximates the flow of information, which is important to control for in light the hypotheses we test. We follow Diether, Lee, and Werner (2009), and calculate price volatility (pvolt) by taking the difference between the daily high price and the daily low price (both from CRSP) and dividing the difference by the daily high price. If short sellers are informed, as the literature suggests, then the level of short selling will be a function of the flow of information into the market. Furthermore, we calculate return volatility (rvolt) as the standard deviation of daily returns from day t - 10 to t, where day t is the current trading day (Lamoureux and Lastrapes, 1990).

Table 1 presents sample descriptive statistics. Panel A of Table 1 reports the firm characteristics of downgraded, upgraded, and affirmed ratings during the entire sample time period, while Panel B reports firm characteristics for downgraded, upgraded, and affirmed, rated firms on the day of the rating announcement. The primary question posed in this study is: does short-selling activity increase prior to IFS ratings? As such it should be noted the mean short ratios of 18.04, 16.12, and 19.26 percent for downgraded, upgraded, and affirmed rated firms suggest that, on average, approximately 18-19 percent of daily trade volume is made up from short sales. These figures are consistent with findings in Diether, Lee, and Werner (2009) and Blau, Van Ness, and Wade (2008). Panel B shows that downgraded firms trade at an average price of $41.27 on the day of the downgrade with an average negative return of -0.0128, whereas upgraded and affirmed rated firms have positive returns on the day of the ratings announcement. It should be noted that on the downgrade day the short ratio increased to 27 percent of trading volume, while in upgraded and affirmed rated firms short selling realized little change in short selling activity.

EMPIRICAL METHODOLOGY

We test the impact of ratings using both univariate and multivariate tests. Furthermore, we test the relation between short selling and the degree of opaqueness in insurers' liabilities and assets around ratings announcements. We conduct a standard event study using an 8-day window around rating announcements. We report market-adjusted returns, which are calculated using the daily (CRSP) raw returns less the equally weighted CRSP index return. We compute the average market adjusted returns for each and run a (cross-sectional) pairwise f-test of changes in the mean.

Our univariate short-selling event study methodology follows several studies that examine trading activity around particular events (Lakonishok and Vermaelen, 1986; Koski and Scruggs, 1998; Sias, 2004) using the following equations:

[standardized short ratio.sub.i,t] = [short ratio.sub.i,t] - [bar.[short ratio.sub.i]]/[sigma]([short ratio.sub.i]) (1)

[standardized short ratio.sub.i,t] = [short turnover.sub.i,t] - [bar.[short turnover.sub.i]]/ [sigma] [sigma]([short turnover.sub.i]) (2)

Specifically, we divide the difference between (1) the trading activity measure on day t for each stock i and (2) the sample period mean of this measure for this stock by (3) the sample period standard deviation of the measure for the stock. The procedure allows for a standardized measure that is similarly distributed across stocks, with a zero mean and a unit variance. This standardization makes shorting activity comparable across stocks with different trading volumes. Further, this standardization normalizes the distribution of both measures of shorting activity. If short selling is abnormally high, then standardized short selling should be significantly different from zero. If more shorting systematically contributes to greater price efficiency, stock prices should deviate less from a random walk (Boehmer, Jones, and Zhang, 2008). Similar to the market-adjusted returns analysis, we compute the average standardized short ratio and standardized short turnover for each and calculate a (cross-sectional) pairwise f-test of changes in the mean.

Our multivariate analysis is performed with panel data models that include both stock and day effects. We use a Hausman specification test to compare the fixed effects versus random effects specifications under the null hypothesis that the individual effects are uncorrelated with the other regressors in the model (Hausman, 1978). The null that random effects exist is rejected in all models and accordingly we estimate the following equations while controlling for both stock and day fixed effects. Recognizing the need to control for other factors that influence the level of short-selling activity, we therefore estimate the following equations using the panel data fixed effect models:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (3)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (4)

The dependent variable is the short selling measure (SSR-standardized short ratio or SSTO-standardized short turnover) from days t - 3 to t - 1. Following Diether, Lee, and Werner (2009) we include firm size ([size.sub.,i,t]), turnover ([Vto.sub.i,t-3,t-1]), price volatility ([pvolt.sub.i,t-3,t-1]), return volatility ([rvolt.sub.i,t-3, t-1]), and contemporaneous market-adjusted returns ([ret.sub.i, t-3, t-1]). As mentioned previously, Diether, Lee, and Werner (2009) find that short selling is contrarian; to control for the contrarian behavior of short sellers, we include the cumulative contemporaneous return ([ret.sub.i, t-3, t-1]). A lagged dependent variable ([SSR.sub.i, t-8, t-4] and [SSTO.sub.i, t-8, t-4] is also included to control for serial correlation in short-sale volume. The variable of interest is [Rating.sub.t], which is an indicator variable equal to one on the IFS downgrade (Down), upgrade (Up), or affirm (Affirm) announcement day, and zero otherwise. If short sellers can anticipate unfavorable ratings changes, then we expect the estimate for Down to be significantly positive.

It has been argued that the insurance industry is more opaque than other industries (Morgan, 2002). In addition, insurance firms vary in their level of opaqueness as their liability and asset structure, while focusing in different lines of insurance business that vary in the level of uncertainty (Zhang, Cox, and Van Ness, 2009). In testing the sophisticated trading hypothesis, we further extend our regression analysis by examining the relation between abnormal short selling and the degree of opaqueness of downgraded insurers as follows:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (5)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (6)

Similar to Equations (3) and (4) we estimate Equations (5) and (6) while controlling for both stock and day fixed effects. Qualitatively similar results are obtained when we estimate standard errors that cluster by both stock and day (Thompson, 2006). The variable specifications are the same as in Equations (3) and (4), while we include the opaqueness measures for both liabilities (Lopaque) and assets (Aopaque) measured by Zhang, Cox, and Van Ness (2009) and Baranoff and Sager (2003). Again, the variables of interest include the indicator variable Rating, which is an indicator variable equal to one on the IFS downgrade (Down), upgrade (Up), or affirm (Affirm) announcement day, and zero otherwise. We also interact the opaqueness variables with the rating indicator variable ([Lopaque.sub.t] x [Rating.sub.t]; [Aopaque.sub.t] x [Rating.sub.t)] to determine the relation of the degree of opaqueness of the firm's liabilities or assets to short selling around IFS rating announcements.

RESULTS

We begin our analysis by examining IFS rating announcements using both univariate and multivariate tests. We also test the relation the opaqueness in insurer's lines of business or assets to short selling around IFS rating announcements.

Short Selling Around IFS Ratings Announcements

Table 2 presents the results of a standard event study using an 8-day window around downgrade (Panel A), upgrade (Panel B) and affirmed (Panel C) IFS ratings. Column [1] shows market-adjusted returns, which are calculated using the daily (CRSP) raw returns less the equally weighted CRSP index return. Consistent with Plalek and Eckles (2010), we find that returns begin to adjust in the 2 days prior to the ratings downgrade. As expected, daily returns are significantly negative (positive) on the event day and continue to remain negative (positive) for 3 (2) days after the downgrade (upgrade). These results suggest that (1) observed downgrades (upgrades) negatively (positively) affect the company's stock price and (2) the stock price begins to adjust before the ratings change is publicly observed. Our findings confirm the findings in Halek and Eckles that prices begin to adjust before the ratings change is publicly observed, suggesting that some investors either predict the downgrade or somehow acquire private information that the ratings will be revised downward. In columns [2] through [5] we examine the short selling surrounding IFS ratings. In columns [2] and [4] we report the short ratio and short turnover, and in columns [3] and [5] the standardized short measures.

The results in Panel A, columns [2] through [5] show that short selling begins to increase a day prior to downgrades for both short selling standardized measures. These initial univariate results affirmatively answer the question whether short sellers can anticipate unfavorable ratings changes, lending initial support for the informational advantage hypothesis and the sophisticated trading hypothesis. On day t -1, both standardized short selling measures are larger than on any other day in the predowngrade period. In economic terms, the standardized short ratio suggests that, on average, short selling increases more than one-half of one standard deviation on day t - 1. Similarly, the standardized short turnover suggests that short selling increases nearly 0.30 standard deviations in the day prior to a ratings downgrade. Rating agencies have likely conducted the analysis before day t - 1, so observing the highest amount of short selling the day before the ratings downgrade during the predowngrade period provides us with further confidence that short sellers have acquired private information about the upcoming ratings change before the information has become publicly available. Table 2 also shows that the short selling spikes on day t and is abnormally high on the day after the downgrade. Again, in economic terms, the short ratio (turnover) increases approximately two-thirds (one-half) of one standard deviation, on average, on the day of the ratings downgrade. These results suggest that short sellers are partially responsible for the downward price response documented in Halek and Eckles (2010).

Panels B and C of Table 2 report the results for upgrades and affirm announcements. Consistent with the idea that short selling contains information about upcoming rating changes, we find that both the short ratio and short turnover are unusually low in the days prior to upgrades (Panel B). We note, however, that the economic magnitude of abnormally low preupgrade short selling is marginal compared to our results in Panel A. For instance, on day t - 1, the short ratio (turnover) is only 0.06 (0.03) standard deviations below mean level of short selling. However, we still observe some statistical significance in columns [3] and [4]. Panel C shows that short selling is relatively close to zero during the period surrounding affirm announcements.

Table 3 reports the regression results from estimating Equations (3) and (4) using the standardized short ratio and standardized short turnover, as the dependent variable, respectively. We find turnover, price volatility, and lagged shorting activity are positively related to short selling. Similar to Diether, Lee, and Werner (2009), we also find short sellers are contrarian in contemporaneous returns as the estimate for [[beta].sub.2] is significantly positive. Further, we still observe abnormally high short selling prior to downgrades. For instance, the indicator variable Down produces a negative and significant estimate in columns [1] and [4]. The magnitude of the coefficient suggests that during the 3 days prior to downgrades, short selling is approximately 0.30 standard deviations above the mean. Columns [2] and [5] show that the estimates for the indicator variable Up are negative but only marginally significant. These results are quantitative and statistically similar in columns [5]--[8] with standardized short turnover being specified as the dependent variable. Further, the economic magnitude of these coefficients is less than half of the magnitude of the coefficients for the variable Down. We do not find that the indicator variable Affirm produces estimates that are reliably different from zero.

The results in Tables 2 and 3 suggest that short sellers are able to successfully anticipate ratings downgrades, and to a lesser extent, upgrades, as we find abnormal high (low) short selling of insurance stocks on the day prior to IFS ratings downgrade (upgrade). (3) Further, the results in Tables 2 and 3 support both the informational advantage hypothesis and the sophisticated trading hypothesis. While short selling prior to unfavorable ratings changes is, at a minimum, consistent with what we would expect to see if information was easily evaluated by market participants, we are left to interpret our results as short sellers are sophisticated traders specifically around IFS rating downgrades. (4)

In addition to being statistically significant, the relation between rating events and short selling is also economically meaningful. For example, the coefficient for Down in column [1] suggests that preevent short selling is 0.321 standard deviations higher than normal. Similar results are found in column [4]. In column [2], the coefficient for Up suggests that preevent short-selling activity is 0.126 standard deviations lower than normal--after controlling for several independent variables.

As mentioned earlier, we focus our attention on the effect of the opaque variables to distinguish between the informational advantage and sophisticated trading hypotheses. Table 4 presents the descriptive statistics for the components of the opaqueness measures calculated following Zhang, Cox, and Van Ness (2009) and Baranoff and Sager (2003). In the first two rows of the table, we find that the mean for Lopaque is 47 percent while the mean for Aopaque is approximately 24 percent. In Table 4, we also provide the mean amounts of premiums written within each line of business.

Table 5 presents the results for Equations (5) and (6). In general, the coefficients on the control variables are similar in sign and magnitude as those in Table 3. Interestingly, we find that the coefficient estimates for Lopaque and Aopaque are significantly negative in both dependent variable specifications and are not statically different. We interpret these results to indicate that firm opaqueness in liabilities and assets negatively affects the level of short-selling activity. However, this result does not lead us to reject the informational advantage hypothesis as it stands.

The variables of interest are the interaction variables, which directly distinguish between the informational advantage hypothesis and the sophisticated trading hypothesis. We interact the opaqueness variables and the dummy variable ([Lopaque.sub.t] x [Down.sub.t]; [Aopaque.sub.t] x [Down.sub.t]) to determine the effect of opaqueness on short selling around an IFS rating downgrade. Columns [1], [4], [5], and [8] show that the interaction estimates are negative and significant--suggesting that pre-downgrade shorting activity is driven by the most transparent (i.e., least opaque) insurers' stocks. This finding makes the case for the informational advantage hypothesis less compelling and suggests that while short sellers have an unusual ability to predict IFS rating downgrades, this is mostly true for insurers with the most transparent assets and liabilities. If short sellers are trading on private information then shorting prior to downgrades should be independent of a firm's opaqueness. Showing that predowngrade short selling is driven by the most transparent firms suggests that predowngrade short selling is better explained by short sellers' sophisticated ability to predict downgrades as opposed to short sellers' ability to trade on private information.

Columns [2], [4], [6], and [8] present the results when we focus on upgrades. In these columns, we do not find significant interaction ([Lopaque.sub.t] x [Up.sub.t]; [Aopaque.sub.t] x [Up.sub.t]) estimates suggesting that the unusually low preupgrade short selling that we observed in the Table 3 are not explained by stocks of insurers with the most balance sheet opacity. Likewise, we do not find significant interaction (Lopaquet x Affirmt; [Aopaque.sub.t] x [Affirm.sub.t]) estimates in columns [3], [4], [7], and [8] when we examine affirm announcements. However, the negative interaction estimates in columns [1], [4], [5], and [8] seem to indicate that the abnormally high predowngrade short selling is greatest in the stocks of insurers with the most transparent balance sheets. To the extent that privately informed short selling is likely to occur in stocks regardless of their opacity, our results seem to suggest that the predowngrade short selling is inversely related to balance sheet opacity. These findings seem to support the sophisticated trading hypothesis instead of the informational advantage hypothesis, leading away from an interpretation that ratings information was privately acquired.

In economic terms, column [1] suggests that a one standard deviation increase in liability opacity reduces the short ratio by nearly 1.73 standard deviations. Similarly, a one standard deviation increase in asset opacity decreases the short ratio by 0.85 standard deviations. Column [5] suggests that a one standard deviation in liability (asset) opacity reduces the short ratio (short turnover) by 2.26 (0.805) standard deviations. Similar results, in terms of economic magnitude are found in the full models in columns [4] and [8].

CONCLUSION

In this study, we examine short-selling activity surrounding IFS ratings announcements in an attempt to determine 1) whether short sellers anticipate ratings changes and 2) whether their anticipation of these changes is driven by privately informed trading. Our results indicate that short sellers successfully anticipate ratings changes as short selling is abnormally high prior to IFS ratings downgrades. Consistent with the idea that short sellers can predict the information contained in these ratings announcements, we also find abnormally low short selling prior to IFS ratings upgrades. Further, short selling prior to announced ratings with no changes (affirmed ratings) remains at relatively normally levels. These findings supports the notion of information flow during the period prior to ratings changes as reported in Halek and Eckles (2010).

In additional tests, we find that short sellers' ability to predict unfavorable rating changes depends on the opacity of the insurers' balance sheet. In particular, we find that the unusually high predowngrade short selling is inversely related to balance sheet opacity. We do not find that the abnormally low preupgrade short selling or the normal preaffirm short selling depends on the balance sheet opacity of insurers. These results seem to suggest that the trading of short sellers prior to ratings changes is not based solely on private information about the upcoming ratings announcement. If short sellers were trading on private information, then short selling levels are likely to be unrelated to the opacity of insurers. On the other hand, if short sellers have an ability to predict these ratings adjustments, as in Flalek and Eckles (2010), then preannouncement short selling is likely to occur in the stocks of insurers with the most transparent balance sheets. Under this argument, our results suggest that the trading of short sellers during the preannouncement period does not seem to be motivated by private information.

DOI: 10.1111/jori.12063

REFERENCES

Aitken, M., A. Frino, M. McCorry, and P. Swan, 1998, Short Sales Are Almost Instantaneously Bad News: Evidence From the Australian Stocks Exchange, Journal of Finance, 53: 2205-2223.

Arnold, T., A. Butler, T. Crack, and Y. Zhang, 2005, The Information Content of Short Interest: A Natural Experiment, Journal of Business, 78: 1307-1335.

Asquith, P., P. A. Pathak, and J. R. Ritter, 2005, Short Interest, Institutional Ownership, and Stock Returns, Journal of Financial Economics, 78: 243-277.

Babbel, D., and C. Merrill, 2005, Real and Illusory Value Creation by Insurance Companies, Journal of Risk and Insurance, 72: 1-21.

Baranoff, E., and T. Sager, 2002, The Relationship Between Asset Risk, Product Risk, and Capital in the Life Insurance Industry, Journal of Banking and Finance, 26:1181-1197.

Blau, B., K. Fuller, and R. Van Ness, 2011, Short Selling Around Dividend Announcements and Ex-dividend Dates, Journal of Corporate Finance, 17: 628639.

Blau, B., and J. Pinegar, 2012, Are Short Sellers Incrementally Informed Prior to Earnings Announcements? Journal of Empirical Finance, 21: 142-155.

Blau, B., R. Van Ness, and C. Wade, 2008, Capitalizing on Catastrophe: Short Selling Insurance Stocks Around Hurricanes Katrina and Rita, Journal of Risk and Insurance, 75: 967-996.

Blau, B., and C. Wade, 2012, Informed or Speculative: Short Selling Analyst Recommendations, Journal of Banking and Finance, 36: 14-25.

Boehmer, E., C. Jones, and X. Zhang, 2008, Which Shorts Are Informed? Journal of Finance, 63: 491-527.

Boehmer, E., and J. Wu, 2013, Short Selling and the Price Discovery Process, Review of Financial Studies, 26: 287-322.

Chakrabarty, B., and A. Shkilko, 2013, Information Transfers and Learning in Financial Markets: Evidence From Short Selling Around Insider Sales, Journal of Banking and Finance, 37: 1560-1572.

Christophe, S., M. Ferri, and J. Angel, 2004, Short Selling Prior to Earnings Announcements, Journal of Finance, 59: 1845-1875.

Christophe, S., M. Ferri, and J. Hsieh, 2010, Informed Trading Before Analyst Downgrades: Evidence From Short Sellers, Journal of Financial Economics, 95: 85-106.

Colquitt, L., R. Hoyt, and K. McCullough, 2006, The Impact of Asbestos and Environmental Reserves Increases on Shareholder Wealth, North American Actuarial Journal, 10: 17-31.

Cummins, J., and P. Danzon, 1997, Price, Financial Quality, and Capital Flows in Insurance Markets, Journal of Financial Intermediation, 6: 3-38.

Daske, H., A. Richardson, and I. Tuna, 2005, Do Short Sale Transactions Precede Bad News Events? Working Paper, University of Pennsylvania.

Dechow, P., A. Hutton, L. Meulbroek, and R. Sloan, 2001, Short-Sellers, Fundamental Analysis, and Stocks Returns, Journal of Financial Economics, 61(1): 77-106.

Desai, H., K. Ramesh, R. Thiagarajan, and B. Balachandran, 2002, An Investigation of the Information Role of Short Interest in the NASDAQ Market, Journal of Finance, 52: 2263-2287.

Diamond, D., and R. Verrecchia, 1987, Constraints on Short Selling and Asset Price Adjustment to Private Information, Journal of Financial Economics, 18: 277-312.

Diether, K., K. Lee, and I. Werner, 2009, Short-Sale Strategies and Return Predictability, Review of Financial Studies, 22: 575-607.

Doherty, N., and R. Phillips, 2002, Keeping up With the Joneses: Changing Rating Standards and the Buildup of Capital by U.S. Property-Liability Insurers, Journal of Financial Services Research, 21: 55-67.

Eckles, D., and M. Halek, 2011, Determinants of Abnormal Reactions to Ratings Downgrades. Working Paper, University of Wisconsin-Madison.

Engelberg, J., A. Reed, and M. Ringgenberg, 2012, How Are Shorts Informed? Short Sellers, News, and Information Processing? Journal of Financial Economics, 105: 260278.

Flannery, M., S. Kwan, and M. Nimalendran, 2004, Market Evidence on the Opaqueness of Banking Firm's Assets, Journal of Financial Economics, 71: 419460.

Halek, M., and D. Eckles, 2010, Effects of Analysts' Ratings on Insurer Stocks Returns: Evidence of Asymmetric Responses, Journal of Risk and Insurance, 77: 801-827.

Hausman, J., 1978, Specification Tests in Econometrics, Econometrica, 46: 1251-1271.

Khan, M., and H. Lu, 2008, Do Short Sellers Front-Run Insider Sales? Working Paper, MIT Sloan School.

Koski, J., and J. Scruggs, 1998, Who Trades Around the Ex-Dividend Day: Evidence From NYSE Audit File Data? Financial Management, 27: 58-72.

Lakonishok, J., and T. Vermaelen, 1986, Tax-Induced Trading Around Ex-Dividend Days, Journal of Financial Economics, 3: 287-319.

Lamoureux, C, and W. Lastrapes, 1990, Heteroskedasticity in Stocks Return Data: Volume Versus GARCH Effects, Journal of Finance, 45: 221-229.

Liu, T., and J. Wu, 2013, Merger Arbitrage Short Selling and Price Pressure, working Paper, University of Georgia.

Morgan, D. P., 2002, Rating Banks: Risk and Uncertainty in an Opaque Industry, American Economic Review, 92: 874-888.

Parekh, R., 2006, Quanta Quits Most Lines After Huge Storm Loss: Lower Rating Cripple Insurer, Business Insurance, 1: 1-12.

Phillips, R., D. Cummins, and F. Allen, 1998, Financial Pricing of Insurance in the Multiple Line Insurance Company, Journal of Risk and Insurance, 65: 597-636.

Polonchek, J., and R. Miller, 1999, Contagion Effects in the Insurance Industry, Journal of Risk and Insurance, 66: 459-475.

Pottier, S., 1998, Life Insurer Financial Stress, Best's Ratings and Financial Ratios, Journal of Risk and Insurance, 65: 275-288.

Pottier, S., and D. Sommer, 1999, Property-Liability Insurer Financial Strength Ratings: Differences Across Ratings Agencies, Journal of Risk and Insurance, 66: 621-642.

Ross, S. 1989, Institutional Markets, Financial Marketing, and Financial Innovation, Journal of Finance, 44: 541-556.

Senchack, A., and L. Starks, 1993, Short Sale Restrictions and Market Reaction to Short-Interest Announcements, Journal of Financial and Quantitative Analysis, 28:177-194.

Sias, R., 2004, Institutional Herding, Review of Financial Studies, 17: 165-206.

Thompson, S. 2006. Simple Formulas for Standard Errors That Cluster by Both Firm and Time. Working Paper, Harvard University.

Zhang, T., L. Cox, and R. Van Ness, 2009, Adverse Selection and the Opaqueness of Insurers, Journal of Risk and Insurance, 76: 295-321.

(1) In other studies, Blau, Fuller, and Van Ness (2011) do not find unusually high short selling prior to changes in dividends. Liu and Wu (2013) do not find abnormally high shorting activity in the acquiring firms' stock prior to merger announcements. These studies support the idea that, in general, short sellers are more reactionary in nature and do not trade prior to informational corporate announcements.

(2) We would like to extend our time period beyond the 2 years; however, the Regulation SHO data are only available from January 2005 to the beginning of 2007, so we are restricted in our time period.

(3) As a robustness check to the specification of the preannouncement short-selling time period, we conducted a multivariate analysis using t - 5, t - 1, and t - 1 windows and observed quantitatively and statistically similar results to that of the t - 3, t - 1 window.

(4) In unreported tests, we replicated our analysis using a modifier used by A.M. Best that creates a watch list. We do not find that the indicator variable capturing "watch list announcements" affects the level of preevent short selling. These results are available upon request.

Chip Wade is at the Mississippi State University, 312 McCool Hall, MS 39762. Wade can be contacted via e-mail: CWade@cobilan.msstate.edu. Andre Liebenberg is at the Department of Finance, University of Mississippi, Oxford, MS. Benjamin M. Blau is at the Department of Economics and Finance, Utah State University, Logan, UT.

Table 1 Summary Statistics Downgrade Upgrade Firms Mean Median Firms Mean Median Panel A: Firm Characteristics Over the Entire Sample Time Period Price 25 37.41 32.3559 14 32.26 28.61 ret 25 0.0007 0.0006 14 0.0017 0.0014 Size 25 12,260,546 1,057,706 14 11,180,312 1,042,512 rvolt 25 0.0154 0.0152 14 0.0112 0.0087 pvolt 25 0.0228 0.0226 14 0.0158 0.0149 Vto 25 0.0049 0.0042 14 0.0038 0.0031 STO 25 0.0019 0.0018 14 0.0023 0.0021 SR 25 0.1804 0.1856 14 0.1612 0.1687 Panel B: Firm Characteristics on Rating Day Price 25 41.2719 38.665 14 36.35 32.31 ret 25 -0.0128 -0.0099 14 0.0102 0.0087 Size 25 12,585,986 1,240,915 14 11,201,568 1,053,588 rvolt 25 0.0166 0.016 14 0.0121 0.0079 pvolt 25 0.1571 0.1328 14 0.0997 0.0978 Vto 25 0.0057 0.0039 14 0.0043 0.0039 STO 25 0.0092 0.0087 14 0.0048 0.0041 SR 25 0.2752 0.245 14 0.1784 0.1741 Affirmed Firms Mean Median Panel A: Firm Characteristics Over the Entire Sample Time Period Price 126 36.23 34.57 ret 126 0.0004 0.0005 Size 126 10,231,421 1,112,467 rvolt 126 0.0101 0.0092 pvolt 126 0.0201 0.0203 Vto 126 0.0045 0.0041 STO 126 0.0013 0.0012 SR 126 0.1926 0.1875 Panel B: Firm Characteristics on Rating Day Price 126 36.82 34.29 ret 126 0.0005 0.0007 Size 126 10,897,456 1,099,236 rvolt 126 0.0122 0.0089 pvolt 126 0.0312 0.0332 Vto 126 0.0052 0.0056 STO 126 0.0018 0.0016 SR 126 0.1897 0.1845 Note: The table shows summary statistics of the sample used in the analysis. Panel A reports the firm characteristics over the entire sample period (January 2005-December 2006) for downgraded, upgraded, and affirmed rated firms. Panel B reports firm characteristics for downgraded, upgraded and affirm rated firms on the day of the rating announcement, price is the average firm share price, ret is the CRSP market-adjusted return, and Size is the CRSP market capitalization, rvolt is the return volatility calculated as the standard deviation of the daily returns from day t - 10 to day t, where day t is the current trading day. pvvolt is the price volatility obtained by taking the difference between the daily high price and the daily low price divided by the daily high price. Turnover (Vto) is the trade volume divided by the shares outstanding while the short turnover (STO) is the short volume divided by the shares outstanding. Short ratio (SR) is the short volume divided by the total volume. Table 2 Short Selling Around A.M. Best Rating Changes return SR SSR STO SSTO [1] [2] [3] [4] [5] Panel A: Downgrade Ratings Returns t - 8, t - 4 0.0093 * 0.2031 0.3977 ** 0.0008 0.2212 t - 3 -0.0036 0.2332 0.3291 0.0006 0.0793 t - 2 -0.0109 ** 0.2106 0.3944 0.0006 0.0963 t - 1 -0.0121 ** 0.2412 0.5122 ** 0.0054 0.2915 ** Event day -0.0128 ** 0.2752 0.6612 ** 0.0092 0.5009 ** t + 1 -0.0039 * 0.2062 0.2433 0.0063 0.3764 ** t + 2 -0.0071 ** 0.2069 0.1293 0.001 0.1534 t + 3 -0.005 0.1999 0.1857 0.0042 0.1692 t + 4, t + 8 0.0009 0.1855 0.3012 0.0017 0.0154 Panel B: Upgrade Ratings Returns t - 8, t - 4 0.0005 0.1856 -0.0142 *** 0.0013 -0.0135 * t - 3 -0.0003 * 0.2211 -0.0541 *** 0.0012 -0.0081 ** t - 2 0.0006 0.2301 -0.0319 ** 0.0016 -0.0046 *** t - 1 0.0035 ** 0.1987 -0.0628 ** 0.0041 -0.0265 ** Event day 0.0102 *** 0.2514 0.0823 *** 0.0073 0.0677 *** t + 1 0.0041 *** 0.2107 0.0471 *** 0.0056 0.0223 ** t + 2 0.0007 ** 0.2008 -0.0256 0.0018 -0.0183 t + 3 0.0002 0.1795 -0.0498 ** 0.0014 -0.0147 t + 4, t + 8 0.0004 0.1979 -0.0227 0.0021 -0.0032 Panel C: Affirmed Ratings Returns t - 8, t - 4 0.0006 0.164 0.0103 0.0011 0.0106 t - 3 0.0004 * 0.175 0.0154 * 0.0009 0.0109 t - 2 0.0008 0.1564 -0.0201 0.0012 0.0094 t - 1 0.0004 0.1621 0.0184 0.0011 0.0123 Event day 0.0005 0.1584 0.0123 0.0013 0.0134 t + 1 0.0004 0.163 0.0203 0.0018 0.0141 t + 2 0.0004 * 0.168 0.0184 0.0021 0.0098 * t + 3 0.0014 0.1723 0.021 * 0.0016 -0.0138 t + 4, t + 8 0.0003 0.1823 -0.0164 0.0014 0.0097 Note: The table shows a standard event study of market-adjusted returns and short selling around A.M. Best rating changes, with Panels A-C representing downgraded, upgraded, and affirmed ratings, respectively. We obtain ratings changes from A.M. Best data and report the return (CRSP equally weighted return), SR, and STO surrounding rating downgrades, where short turnover (STO) is the short volume divided by the shares outstanding and short ratio (SR) is the short volume divided by the total volume. Tests for significant returns are determined by standard f-statistics testing for differences from zero. We test for the significance in short selling using two different methods (SSR and SSTO). We also standardize short-selling activity by calculating the difference between the short activity for stock i on day t and the mean short activity for stock i (across the sample time period). We then divide the difference by the standard deviation of daily short activity so that each short measure on each day is similarly distributed with a zero mean and a unit variance. t-Statistics testing whether the standardized measure is significantly different than zero (the mean) are obtain. The f-tests test whether the standardized and abnormal measures are significantly different from zero (the mean) ***, **, and * indicate significance at the 0.01, 0.05, and 0.1 levels, respectively. Table 3 Panel Regression Results [SSR.sub.i, t-3, t-1] [1] [2] Intercept 3.048 *** 3.254 ** (0.000) (0.000) [Size.sub.i] -0.165 ** -0.203 ** (0.024) (0.036) [ret.sub.i, t-3, t-1] 11.432 *** 8.387 ** '(0.000) (0.044) [Vto.sub.i, t-3, t-1] 1.254 ** 2.013 * (0.023) (0.065) [pvolt.sub.i, t-3, t-1] 3.263 ** 2.874 * (0.029) (0.087) [rvolt.sub.i, t-3, t-1] -1.895 -1.745 (0.301) (0.152) [SSTO.sub.i, t-8, t-4] [SSR.sub.i, t-8, t-4] 0.285 *** 0.316 *** '(0.000) '(0.000) [Down.sub.t] 0.321 ** (0.029) [Up.sub.t] -0.126 * (0.064) [Affirm.sub.t] [F-Stat.sub.Down = Up] [F-Stat.sub.Down = Affirm] [F-Stat.sub.Up = Affirm] Adj [R.sup.2] 0.312 0.285 Stock FE Yes Yes Day FE Yes Yes Observations 12,475 6,986 [SSR.sub.i, t-3, t-1] [3] [4] Intercept 4.036 ** 3.852 ** (0.000) (0.000) [Size.sub.i] -0.284 -0.265 (0.138) (0.235) [ret.sub.i, t-3, t-1] 7.228 ** 9.318 ** (0.035) (0.048) [Vto.sub.i, t-3, t-1] 1.787 1.547 * (0.234) (0.076) [pvolt.sub.i, t-3, t-1] 3.523 * 3.012 * (0.063) (0.074) [rvolt.sub.i, t-3, t-1] -1.012 -1.458 (0.135) (0.218) [SSTO.sub.i, t-8, t-4] [SSR.sub.i, t-8, t-4] 0.273 ** 0.303 ** (0.036) (0.011) [Down.sub.t] 0.306 ** (0.042) [Up.sub.t] -0.168 (0.109) [Affirm.sub.t] 0.187 0.201 (0.302) (0.247) [F-Stat.sub.Down = Up] 9.62 ** (0.011) [F-Stat.sub.Down = Affirm] 10.23 ** (0.001) [F-Stat.sub.Up = Affirm] 1.27 (0.206) Adj [R.sup.2] 0.331 0.326 Stock FE Yes Yes Day FE Yes Yes Observations 62,874 82,335 [SSTO.sub.i, t-3, t-1] [5] [6] Intercept 2.963 *** 3.247 *** (0.000) (0.000) [Size.sub.i] -0.201 *** -0.268 * '(0.000) (0.069) [ret.sub.i, t-3, t-1] 10.235 ** 9.366 ** (0.022) (0.017) [Vto.sub.i, t-3, t-1] 2.014 ** 1.589 * (0.044) (0.062) [pvolt.sub.i, t-3, t-1] 4.212 ** 4.014 *** (0.042) (0.000) [rvolt.sub.i, t-3, t-1] -2.036 ** -1.985 ** (0.032) (0.019) [SSTO.sub.i, t-8, t-4] 0.358 *** 0.426 *** '(0.000) '(0.000) [SSR.sub.i, t-8, t-4] [Down.sub.t] 0.284 ** (0.031) [Up.sub.t] -0.198 (0.184) [Affirm.sub.t] [F-Stat.sub.Down = Up] [F-Stat.sub.Down = Affirm] [F-Stat.sub.Up = Affirm] Adj [R.sup.2] 0.389 0.396 Stock FE Yes Yes Day FE Yes Yes Observations 12,475 6,986 [SSTO.sub.i, t-3, t-1] [7] [8] Intercept 4.012 ** 3.597 ** (0.000) (0.000) [Size.sub.i] -0.132 -0.263 (0.224) (0.303) [ret.sub.i, t-3, t-1] 5.325 * 7.564 ** (0.059) (0.036) [Vto.sub.i, t-3, t-1] 1.889 * 1.646 * (0.074) (0.051) [pvolt.sub.i, t-3, t-1] 3.523 ** 3.864 ** (0.033) (0.047) [rvolt.sub.i, t-3, t-1] -2.113 -1.897 (0.235) (0.106) [SSTO.sub.i, t-8, t-4] 0.268 ** 0.381 *** (0.023) (0.000) [SSR.sub.i, t-8, t-4] [Down.sub.t] 0.294 ** (0.017) [Up.sub.t] -0.201 (0.149) [Affirm.sub.t] 0.235 0.217 (0.214) (0.189) [F-Stat.sub.Down = Up] 8.47 ** (0.036) [F-Stat.sub.Down = Affirm] 7.66 ** (0.024) [F-Stat.sub.Up = Affirm] 2.13 (0.162) Adj [R.sup.2] 0.314 0.358 Stock FE Yes Yes Day FE Yes Yes Observations 62,874 82,335 Note: The table presents the panel regression results from estimating the following equations where the dependent variables are measures of short-selling activity from days t-3 to t-1. Short-selling measures are the standardized short ratio ([SSR.sub.i, t-3, t-1]) and standardized short turnover ([SSTO.sub.i, t-3, t-1]). [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] The independent variables include contemporaneous share turnover ([Vto.sub.i, t-3-, t-1]), return volatility ([rvolt.sub.i, t-3, t-1]), and price volatility ([pvolt.sub.i, t-3, t-1]). We also include a lagged dependent variable to control for serial correlation ([SSTO.sub,.t-8, t-4] and [SRO,.sub.t-8, t-4]) and the contemporaneous market-adjusted return ([ret.sub.i, t-3, t-1]). Size is the CRSP market capitalization. The variable of interest is [Rating.sub.t], which is an indicator variable equal to one on the IFSR downgrade (Down), upgrade (Up), or affirm (Affirm) announcement day. A Hausman test reveals observed differences across stocks and days so we report two-way fixed effects estimates. In columns [4] and [8] we compare the estimates of downgrades, upgrades, and affirmed using a standard F-test. P-values are reported in parentheses ***, **, and * indicate significance at the 0.01, 0.05, and 0.1 levels, respectively. Table 4 Summary Statistics Variable Mean Std. Dev Min Max Aopaque (%) 25.425 25.965 2.08 28.92 Lopaque (%) 47.49 59.48 43.33 78.205 Components P/C-Liabilities-Opaque (%) 24.83 56.15 0 95.63 Total P/C premium ($1,000) 2,458 4,589 0 25,362 Aircraft premium 20 77 0 662 Automobile premium 43 153 0 1,175 Fidelity premium 8 33 0 184 Medical malpractice premium 30 94 0 566 Surety premium 27 95 0 793 Workers compensation premium 207 495 0 3,107 Other liability premium 120 390 0 3,076 P/C-Assets-Opaque (%) 34.05 41.01 4.16 21.66 Total P/C assets ($1,000) 1,671 6,044 24 94,034 Real estate 40 121 0 853 Mortgage 16 61 0 434 Premium notes 1 2 0 21 Premium receivables 195 360 0 2,032 Other investments 287 1,889 1 16,835 Reinsurance recoverable 30 46 0 193 L/H-Liabilities-Opaque (%) 70.15 62.81 86.66 60.78 Total L/H premium ($1,000) 5,441 6,718 15 39,351 Accident and health (Only) premium 3,817 4,220 13 23,920 L/H-Assets-Opaque (%) 16.8 10.92 0 36.18 Total L/H assets ($1,000) 2,967 8,302 84 12,338 Bonds 389 595 0 2,542 Real estate 22 75 0 549 Mortgage 24 51 0 332 Premium notes 0.0008 0.0041 0 0.0312 Premium receivables 5 41 0 312 Other investments 44 76 0 295 Reinsurance recoverable 17 69 0 434 Note: The table shows summary statistics for the components of the opaqueness measures. Table 5 Panel Regression Results and F-Statistics [SSR.sub.i, t-3, t-1] [1] [2] Panel A: Panel Regression Results Intercept 2.989 *** 3.001 *** (0.000) (0.000) [Size.sub.t] -0.134 *** -0.139 * (0.000) (0.067) [ret.sub.i, t-3, t-1] 8.542 *** 8.621 ** (0.000) (0.043) [Vto.sub.i, t-3, t-1] 0.311 0.312 (0.623) (0.521) [pvolt.sub.i, t-3, t-1] 2.126 ** 2.087 ** (0.026) (0.018) [rvolt.sub.i, t-3, t-1] -0.523 -0.514 (0.312) (0.306) [SSTO.sub.i, t-8, t-4] [SSR.sub.i, t-8, t-4] 0.221 *** 0.264 ** (0.000) (0.031) [Lopaque.sub.t] -0.021 *** -0.023 *** (0.032) (0.028) [Aopaque.sub.t] -0.021 ** -0.023 ** (0.037) (0.042) [Down.sub.t] 0.062 ** (0.016) [Up.sub.t] 0.025 (0.102) [Affirm.sub.t] [Lopaque.sub.t] x [Down.sub.t] -0.029 ** (0.026) [Aopaque.sub.t] x [Down.sub.t] -0.033 ** (0.034) [Lopatjue.sub.t] x [Up.sub.t] -0.009 (0.301) [Aopaque.sub.t] x [Up.sub.t] -0.013 (0.132) [Lopaque.sub.t] x [Affirm.sub.t] [Aopaque.sub.t] x [Affirm.sub.t] Adj [R.sup.2] 0.387 0.325 Stock FE Yes Yes Day FE Yes Yes Observations 12,475 6,986 Panel B: F-Statistic Results [F-Stat.sub.Lopaque=Aopaque] [F-Stat.sub.Down=Up] [F-Stat.sub.Down=Affirm] [F-Stat.sub.U=Affirm] [F-Stat.LopaquexDown=LopaquexUp] [F-Stat.sub.LopaquexDown=LopaquexAffirm] [F-Stat.sub.LopaquexUp=LopaquexAffirm] [F-Stat.sub.AopaquexDown=AopaquexUp] [F-Stat.sub.AopaquexDown=AopaquexAffirm] [F-Stat.sub.AopaquexUp=AopaquexAffirm Observations [SSR.sub.i, t-3, t-1] [3] [4] Panel A: Panel Regression Results Intercept 3.002 *** 2.874 *** (0.000) (0.000) [Size.sub.t] -0.142 ** -0.154 ** (0.048) (0.031) [ret.sub.i, t-3, t-1] 9.087 ** 8.241 ** (0.036) (0.025) [Vto.sub.i, t-3, t-1] 0.331 0.298 (0.447) (0.216) [pvolt.sub.i, t-3, t-1] 2.158 * 2.203 * (0.073) (0.055) [rvolt.sub.i, t-3, t-1] -0.547 -0.507 (0.289) (0.158) [SSTO.sub.i, t-8, t-4] [SSR.sub.i, t-8, t-4] 0.258 * 0.227 * (0.072) (0.053) [Lopaque.sub.t] -0.013 * -0.031 ** (0.059) (0.026) [Aopaque.sub.t] -0.028 ** -0.033 * (0.031) (0.056) [Down.sub.t] 0.059 ** (0.022) [Up.sub.t] 0.031 (0.127) [Affirm.sub.t] 0.035 0.029 (0.163) (0.206) [Lopaque.sub.t] x [Down.sub.t] -0.038 ** (0.014) [Aopaque.sub.t] x [Down.sub.t] -0.029 ** (0.021) [Lopatjue.sub.t] x [Up.sub.t] -0.012 (0.156) [Aopaque.sub.t] x [Up.sub.t] -0.024 (0.158) [Lopaque.sub.t] x [Affirm.sub.t] 0.029 0.019 (0.122) (0.143) [Aopaque.sub.t] x [Affirm.sub.t] 0.023 0.029 (0.231) (0.158) Adj [R.sup.2] 0.386 0.356 Stock FE Yes Yes Day FE Yes Yes Observations 62,874 82,335 Panel B: F-Statistic Results [F-Stat.sub.Lopaque=Aopaque] 1.43 (0.231) [F-Stat.sub.Down=Up] 6.63 ** (0.016) [F-Stat.sub.Down=Affirm] 8.92 *** (0.000) [F-Stat.sub.U=Affirm] 0.13 (0.339) [F-Stat.LopaquexDown=LopaquexUp] 7.42 ** (0.034) [F-Stat.sub.LopaquexDown=LopaquexAffirm] 6.98 ** (0.018) [F-Stat.sub.LopaquexUp=LopaquexAffirm] 0.98 (0.313) [F-Stat.sub.AopaquexDown=AopaquexUp] 8.78 ** (0.027) [F-Stat.sub.AopaquexDown=AopaquexAffirm] 9.43 *** (0.000) [F-Stat.sub.AopaquexUp=AopaquexAffirm 1.27 (0.296) Observations 82,335 [SSTO.sub.i, t-3, t-1] [5] [6] Panel A: Panel Regression Results Intercept 2.963 *** 2.961 *** (0.000) (0.000) [Size.sub.t] -0.201 ** -0.218 * (0.035) (0.074) [ret.sub.i, t-3, t-1] 10.036 ** 9.997 ** (0.031) (0.027) [Vto.sub.i, t-3, t-1] 0.401 0.413 (0.412) (0.387) [pvolt.sub.i, t-3, t-1] 3.213 ** 3.156 * (0.041) (0.073) [rvolt.sub.i, t-3, t-1] -0.841 *** -0.856 ** (0.000) (0.034) [SSTO.sub.i, t-8, t-4] 0.384 ** 0.387 ** (0.043) (0.039) [SSR.sub.i, t-8, t-4] [Lopaque.sub.t] -0.024 ** -0.025 ** (0.023) (0.018) [Aopaque.sub.t] -0.036 ** -0.032 ** (0.026) (0.018) [Down.sub.t] 0.071 ** (0.034) [Up.sub.t] 0.047 (0.316) [Affirm.sub.t] [Lopaque.sub.t] x [Down.sub.t] -0.038 ** (0.009) [Aopaque.sub.t] x [Down.sub.t] -0.031 ** (0.041) [Lopatjue.sub.t] x [Up.sub.t] 0.014 (0.206) [Aopaque.sub.t] x [Up.sub.t] 0.023 (0.199) [Lopaque.sub.t] x [Affirm.sub.t] [Aopaque.sub.t] x [Affirm.sub.t] Adj [R.sup.2] 0.399 0.401 Stock FE Yes Yes Day FE Yes Yes Observations 12,475 6,986 Panel B: F-Statistic Results [F-Stat.sub.Lopaque=Aopaque] [F-Stat.sub.Down=Up] [F-Stat.sub.Down=Affirm] [F-Stat.sub.U=Affirm] [F-Stat.LopaquexDown=LopaquexUp] [F-Stat.sub.LopaquexDown=LopaquexAffirm] [F-Stat.sub.LopaquexUp=LopaquexAffirm] [F-Stat.sub.AopaquexDown=AopaquexUp] [F-Stat.sub.AopaquexDown=AopaquexAffirm] [F-Stat.sub.AopaquexUp=AopaquexAffirm Observations [SSTO.sub.i, t-3, t-1] [7] [8] Panel A: Panel Regression Results Intercept 2.945 *** 2.844 *** (0.000) (0.000) [Size.sub.t] -0.208 * -0.226 * (0.059) (0.062) [ret.sub.i, t-3, t-1] 10.149 * 9.889 * (0.064) (0.071) [Vto.sub.i, t-3, t-1] 0.409 0.421 (0.436) (0.234) [pvolt.sub.i, t-3, t-1] 3.233 * 3.207 * (0.068) (0.076) [rvolt.sub.i, t-3, t-1] -0.784 -0.776 (0.269) (0.189) [SSTO.sub.i, t-8, t-4] 0.365 ** 0.399 ** (0.047) (0.038) [SSR.sub.i, t-8, t-4] [Lopaque.sub.t] -0.015 ** -0.032 ** (0.027) (0.036) [Aopaque.sub.t] -0.029 ** -0.039 * (0.034) (0.064) [Down.sub.t] 0.084 ** (0.022) [Up.sub.t] 0.056 (0.204) [Affirm.sub.t] 0.038 0.041 (0.217) (0.168) [Lopaque.sub.t] x [Down.sub.t] -0.046 ** (0.023) [Aopaque.sub.t] x [Down.sub.t] -0.038 ** (0.022) [Lopatjue.sub.t] x [Up.sub.t] 0.027 (0.213) [Aopaque.sub.t] x [Up.sub.t] 0.031 (0.204) [Lopaque.sub.t] x [Affirm.sub.t] 0.018 0.023 (0.184) (0.177) [Aopaque.sub.t] x [Affirm.sub.t] 0.028 0.034 (0.152) (0.206) Adj [R.sup.2] 0.419 0.406) Stock FE Yes Yes Day FE Yes Yes Observations 62,874 82,335 Panel B: F-Statistic Results [F-Stat.sub.Lopaque=Aopaque] 2.31 (0.297) [F-Stat.sub.Down=Up] 7.74 ** (0.028) [F-Stat.sub.Down=Affirm] 9.74 *** (0.000) [F-Stat.sub.U=Affirm] 0.058 (0.313) [F-Stat.LopaquexDown=LopaquexUp] 6.42 ** (0.036) [F-Stat.sub.LopaquexDown=LopaquexAffirm] 8.59 ** (0.011) [F-Stat.sub.LopaquexUp=LopaquexAffirm] 1.58 (0.196) [F-Stat.sub.AopaquexDown=AopaquexUp] 6.19 ** (0.044) [F-Stat.sub.AopaquexDown=AopaquexAffirm] 9.56 *** (0.000) [F-Stat.sub.AopaquexUp=AopaquexAffirm 0.087 (0.307) Observations 82,335 Note: Panel A presents the panel regression results from estimating the following equations where the dependent variables are measures of short-selling activity from days t-3 to t-1. Short-selling measures are the standardized short ratio ([SSR.sub.i, t-3, t-1]) and standardized short turnover ([SSTO.sub.i, t-3, t-1]). [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] The independent variables include contemporaneous share turnover ([Vto.sub.i, t-1]), return volatility ([rvol.sub.i, t-3, [t-1]), and price volatility ([pvolt.sub.I, t-3, t-1]). We also include a lagged dependent variable to control for serial correlation ([SSTO,.sub.t-8, t-4]) and [SRO.sub.,t-8, t-4]) and the contemporaneous market-adjusted return ([ref.sub.i, t-3, t-1]). Size is the CRSP market capitalization. The variable of interest is [Rating.sub.t], which is an indicator variable equal to one on the IFSR downgrade (Down), upgrade (Up), or affirm (Affirm) announcement day. Further, the levels of opaqueness in liabilities (Lopaque) and assets (Aopaque) are represented. We also interact the rating dummy variable and the continuous variable foropaqueness in line of business and assets ([Down.sub.t] x [Lopaque.sub.t]; [Down.sub.t] x [Aopaque.sub.t], [Up.sub.t] x [Lopaque.sub.t]; [Up.sub.t] x [Aopaque.sub.t], [Affirm.sub.t], x [Lopaques.ub.t]; [Affirm.sub.t], x [Aopaque.sub.t]) to determine whether the short selling is affected by the opaqueness of the insurer's line of business or assets during the preannouncement period. A Hausman test reveals observed differences across stocks and days so we report two-way fixed effects estimates. Panel B of the table presents the results of the F-statisties when testing the differences in Rating, the levels of opaqueness in liabilities (Lopaque) and assets (Aopaque), and the interaction of these indicator variables. In columns [4] and [8] we compare the estimates of downgrades, upgrades, and affirmed using a standard F-test. P-values are reported in parentheses. ***, **, and * indicate significance at the 0.01,0.05, and 0.1 levels, respectively.

Printer friendly Cite/link Email Feedback | |

Author: | Wade, Chip; Liebenberg, Andre; Blau, Benjamin M. |
---|---|

Publication: | Journal of Risk and Insurance |

Geographic Code: | 1USA |

Date: | Jun 1, 2016 |

Words: | 12158 |

Previous Article: | The pricing of mortgage insurance premiums under systematic and idiosyncratic shocks. |

Next Article: | Making a Market for Acts of God: The Practice of Risk-Trading in the Global Reinsurance Industry. |

Topics: |