Journal of Empirical Finance 25 (2014) 83–94

Contents lists available at ScienceDirect

Journal of Empirical Finance

j ourna l homepage: www.e lsev ie r .com/ locate / jempf in

Firm opacity and financial market information asymmetry☆

Rahul Ravi⁎,1, Youna Hong 2

John Molson School of Business, Concordia University, Canada

a r t i c l e i n f o

☆ We thank Aditya Kaul, Mark Huson, Vikas Mehrotra,HeatherWeir, Sandra Betton, Imants Paeglis andNilanjantwo anonymous referees for the helpful comments and s⁎ Corresponding author.

E-mail addresses: rravi@jmsb.concordia.ca (R. Rav1 Rahul Ravi is an assistant professor in the departm2 Youna Hong is a Ph.D. student in the department o

of Finance, at the John Molson School of Business wh

0927-5398/$ – see front matter © 2013 Elsevier B.V. Ahttp://dx.doi.org/10.1016/j.jempfin.2013.11.007

a b s t r a c t

Article history:Received 27 January 2012Received in revised form 29 October 2013Accepted 22 November 2013Available online 1 December 2013

Information asymmetry could exist between the firm and the investors as well as amonginvestors. If the information asymmetry between the firm and the investors is very high, allinvestors are largely uninformed, so information asymmetry between investors should be low.At the other extreme, if all investors are fully informed about the firm, again the informationasymmetry between investors should be low. This paper finds evidence supporting such anonlinear relationship between firm-to-investor and investor-to-investor information asym-metry. The inter-investor information asymmetry increases, and then declines, as theinformation asymmetry between the firm and the investor increases.

© 2013 Elsevier B.V. All rights reserved.

JEL classification:D82G19M40

Keywords:Information asymmetryFirm opacityDisclosure quality

1. Introduction

Information asymmetry between the firm and the investor affects the functioning of an efficient capital market (Healy andPalepu, 2001). Considerable resources have been devoted by the regulators to enact and enforce new disclosure policies (such asRegulation Fair Disclosure) aimed at reducing information asymmetry between the firm and the investor. An implicit expectationis that these regulations will reduce information asymmetry among investors in the financial market making it more fair andefficient. This last statement assumes a monotonic relation between the two types of information asymmetries. However, extanttheoretical studies cast some doubt on this assumption.

Some researchers (Diamond, 1985; Hakansson, 1977) find that reduction in firm-to-investor information asymmetry(hereafter referred to as firm-to-investor-IA) leads to reduction in the expected net benefit to investors with private information,thereby lowering their incentive to find the information in the first place. Thus, lower firm-to-investor-IA should imply a lowerrisk of trading with an informationally-endowed trader. This reduces the level of inter-investor-information asymmetry(hereafter referred to as inter-investor-IA). However, other studies (Kandel and Pearson, 2005; Kim and Verrecchia, 1994;Lundholm, 1991) suggest that greater quantities of and better quality information released by a firm provide more material tothose investors who attempt to process public signals in order to create private benefits. If this is the case, then lowering

Barry Scholnick, Connie Smith, Marc Lipson, Ron Masulis, Kumar Venkatraman, Srinivasan Sankaraguruswamy,Basu for comments. I thank Brian Bushee for providing AIMR disclosure data tome.Wewould also like to thankuggestions. We thank IFM2 for financial support.

i), YouNa.Hong13@rotman.utoronto.ca (Y. Hong).ent of finance at the John Molson School of Business, Concordia University, Montreal, Quebec, Canada.

f Finance, at the Rotman School of Business, University of Toronto. She was M.Sc. student in the departmentile working on this paper.

ll rights reserved.

84 R. Ravi, Y. Hong / Journal of Empirical Finance 25 (2014) 83–94

firm-to-investor-IA may result in higher inter-investor-IA. Given the differing views, it is not clear whether a reduction in onewould lead to a decrease or an increase in the other.

Since theoretical arguments alone cannot unambiguously predict the relationship between firm-to-investor-IA andinter-investor-IA, we look towards empirical research for some guidance. Most of the disclosure literature (Brown andHillegeist, 2007; Healy et al., 1999; Heflin et al., 2005; Welker, 1995) find a negative relation between disclosure quality andinformation asymmetry among investors, as proxied by various measures of the bid–ask spread. Brown and Hillegeist (2007)argue that this negative relationship exists because improved disclosure quality acts as a disincentive for investors searching forprivate information, and therefore it reduces the probability of trading with informed traders in the market. This view isconsistent with the first of the two theoretical views outlined above (Diamond, 1985; Hakansson, 1977). However, evidencefrom another set of studies, which includes Lee (1992), Lee et al. (1993), Krinsky and Lee (1996), Huson andMacKinnon (2003),Ke and Ramaligegowda (2005), and Desaia and Savickas (2010), suggests a positive relation between disclosure quality andinformation asymmetry among investors.

Huson and MacKinnon (2003), while studying the effects of focus enhancing spinoffs on inter-investor-IA, find that theinformation asymmetry among investors increases post spinoff. To the extent that a focus-enhancing spinoff leads toreduction in firm-to-investor-IA, their evidence supports the second theoretical view outlined above (Kandel and Pearson,2005; Kim and Verrecchia, 1994; Lundholm, 1991). Desaia and Savickas (2010) draw a similar conclusion from theirevidence that the idiosyncratic volatility for parent firms increased after spinoffs and equity carve-outs. Exploring the postearnings announcement adverse selection cost of trading (a measure of inter-investor information asymmetry), Lee et al.(1993) and Krinsky and Lee (1996) find that, on average, the adverse selection cost component of the bid–ask spreadincreases significantly after earnings announcements. Similar evidence has also been found in studies of post earningsannouncement drift trading strategies. Ke and Ramaligegowda (2005) document that the “transient” institutions thatarbitrage drift can trade on their ability to quickly process public disclosures into tradable private information. Their findingssuggest that differential private information can be created from public disclosures by investors with better processingability. Thus, once again, the extant research leaves us with differing views on the relationship between firm-to-investor-IAand inter-investor-IA.

We attempt to reconcile the two diverging views by proposing a unimodal relation between the two types of informationasymmetries. This proposal draws its intuition from the theoretical model of Kim and Verrecchia (1991). Our contribution here isto provide empirical evidence of a nonlinear and concave relation between the two types of information asymmetry thatpotentially synthesizes seemingly conflicting extant research.

This line of reasoning is primarily motivated by Kim and Verrecchia (1991)'s work on the relation between the precision ofpublic announcement and inter-investor-IA. At one extreme, if the precision of the public announcement is low, disclosure causeslittle expectation revision and thus few opportunities to trade. Consequently, there is little (additional) incentive for investors toacquire private information in anticipation of a public announcement, so that information asymmetry across investors remainssmall. At the other extreme, suppose that the precision of the public announcement is high. Here, the disclosure is of sufficientmagnitude to shock previous beliefs and price equilibrium. Hence, no opportunities are created for investors to trade. Again, thereis little (additional) incentive for investors to acquire private information, and therefore information asymmetry across investorsbecomes small. Between these two extremes, the impact of the public announcement is sufficiently large to create opportunitiesto trade, but it does not eliminate uncertainties to cause all beliefs and price to converge. Thus, information asymmetry acrossinvestors becomes substantial. Therefore, according to this model, information asymmetry across investors increases as theprecision of the public announcement increases up to some point, and steadily decreases thereafter (that is, its behavior isunimodal).

We test this prediction using a sample of over 1000 firms listed on NYSE from January 1993 to December 2008 and findevidence in support of this hypothesis. As firm-to-investor-IA increases, inter-investor-IA increases up to some point, and steadilydecreases thereafter. This result is intuitively appealing. If a firm is completely transparent, all market participants would knoweverything about the firm, hence, the inter-investor-IA should be zero.3 If the firm is completely opaque, all participants areuninformed, hence, the inter-investor-IA should be close to zero. Somewhere between the two extremes, the informationasymmetry between investors attains a maximum.4

The result of this study has implications for a firm's transparency and disclosure-related policies. A marginal increase in thelevel of transparency of a firm with very high firm-to-investor-IA could lead to an increase in its inter-investor-IA. This might leadto reduced liquidity and possibly to a higher cost of capital (Amihud and Mendelson, 1988). The results suggest that increasedfirm transparency might not necessarily be always advantageous to the average investor in the market. Thus, this study adds tothe literature on corporate disclosures policies. Furthermore, by pointing to a nonlinear relationship between firm-to-investor-IAand inter-investor-IA, this study also provides a cautionary note to anyone using microstructure based information asymmetrymeasures to proxy for firm-to-investor-IA.

3 A transparent firm is one with low to nil firm-to-investor-IA. As the level of firm-to-investor-IA declines, firms will become less transparent (or more opaque).This paper uses transparency and firm-to-investor-IA synonymously. Opacity is the antonym of transparency.

4 A caveat is in order here. At the two extremes, the level of inter-investor-IA will be determined through the interplay of search costs associated with obtainingprivate information and the economic value of the obtained private information. Therefore, the level of inter-investor-IA might be non-zero.

85R. Ravi, Y. Hong / Journal of Empirical Finance 25 (2014) 83–94

The remainder of this paper is organized as follows. In Section 2, we present the sample and the methodology used inthis study and Section 3 describes the results. Section 4 concludes the paper with a brief discussion of our study'simplications.

2. Data and methodology

2.1. Data

We start with all firms present in NYSE Trade and Quote (TAQ) database from January 1993 to December 2008. In orderto control for cross-market microstructure effects, we restrict the sample to firms listed on the NYSE. 5 To control for anypossible calendar effects, we further restrict the sample to firms with January to December fiscal year. We also excludeutilities (SIC codes from 49 to 50) and firms from the financial sector (SIC codes from 60 to 68). ADRs or other securities,incorporated outside the US, as well as preferred stocks and other non-common stocks, are excluded.6 All firms involved inany spin-off activity in a given year are excluded from the analysis. To be included in the sample, a firm has to be present inboth TAQ and CRSP for all 12 months of the respective fiscal year. To avoid undue influence from extreme observations,firms with stock prices below $5 or above $500 are excluded. The final sample size ranges from a low of 630 in 1993 to 1022in 2008 (Table 1).

2.1.1. Measure of inter-investor-IAWe use the adverse selection cost component of spread, as suggested by the Glosten and Harris (1988) model, to measure

inter-investor-IA.7 In the Glosten and Harris (1988) model, the adverse selection, inventory-holding, and order-processingcomponents are expressed as a linear function of transaction volume. The model is described as follows:

5 Thecontrol

6 Sec7 We

robust(1988)

8 LeeduringcontemWei (20five-sec

9 Wea non-peliminaquote mthe ask-16, 17,

ΔPt ¼ c0ΔIt þ c1ΔItVt þ z0It þ z1ItVt þ εt : ð1Þ

In this case, It is a trade indicator that equals 1 if the transaction is buyer-initiated, and −1 if it is seller-initiated; Pt is thetransaction price at time t; Vt is the volume traded at time t; and εt captures public information innovations and errors. In thismodel, the adverse-selection component is 2(z0 + z1Vt), and other components (inventory-holding and order-processingcomponents) are measured as 2(c0 + c1Vt). We use the average transaction volume for the stock to obtain an estimate of theadverse-selection component as a percentage of the bid–ask spread:

GH ¼ 2 z0 þ z1V� �

2 c0 þ c1V� �þ 2 z0 þ z1V

� � : ð2Þ

We follow the Lee and Ready (1991) procedure for classifying trades. According to this algorithm, a trade is classified as buyer-(seller-) initiated if the transaction price is closer to the ask (bid) price of the prevailing quote. The quote must be at least onesecond old.8 If the trade is at the exact midpoint of the quote, a ‘tick test’ classifies the trade as buyer- (seller-) initiated if the lastprice change prior to the trade is positive (negative). Since the trade direction is inferred from the available information and notobserved, some assignment error is inevitable. Hence, the resulting order-flow data is an estimate. Nevertheless, as shown by Leeand Radhakrishna (2000) and Odders-White (2000), the Lee and Ready (1991) algorithm is largely accurate; thus, inferencesbased on the estimated order-flow should be reliable.

Several filters are employed to ensure the validity of the TAQ data.9 The first trade of each day is dropped from the analysis,since it usually occurs through a call auction. The TAQ database does not eliminate auto-quotes (passive quotes by secondary

spread decomposition methodologies used in this paper are sensitive to the trading mechanism. Restricting to sample to only NYSE listed firms helpsfor this known issue.urities with CRSP share codes other than 10 or 11 were excluded.have also run the analysis using the adverse selection cost component as proposed by Lin et al. (1995) and Neal and Wheatley (1998). Our results areto the methodology selected. For the sake of brevity, we discuss only the results corresponding to the measure computed using the Glosten and Harrisalgorithm.and Ready (1991) propose a five second quote lag to account for quotes being updated before the trades that triggered them were reported, becausetheir sample period quotes were updated on a computer while trades were entered manually. Bessembinder (2003) finds that by 1998, usingporaneous quotes was preferred than using the five second lagged quotes. Vergote (2005) reports that a two-second lag is optimal, while Piwowar and06) suggest that a one-second lag produces superior trade direction inferences. We have analyzed a subset of our data (1993 through 2003) using theond rule. The results remain unchanged, as compared to those derived from one second matching.drop all trades with a correction indicator other than 0 or 1, and retain only those trades for which the condition is B, J, K, or S. We also drop all trades withositive trade size or price. We omit all trades recorded before opening time or after the closing time of the market. Negative bid–ask spreads areted. To avoid using incorrectly recorded quotes, we eliminate all quotes for which quoted spread is greater than 20% of the quote mid-point, when theid-point is greater than $10 or quoted spread is greater than $2, when the quote mid-point is less than $10. We also eliminate all quotes for which eitheror the bid-quote moves by more than 50%. All trades with conditions A, C, D, N, O, R, or Z are excluded. All quotes with conditions 5, 7, 8, 9, 11, 13, 14, 15,

19, 20, 27, 28, 29 are excluded.

Table 1Sample building.Our base samples are US-incorporated common stocks listed on the NYSE from January 1993 to December 2006. We apply the following filters: Include firms withfiscal year ending in December; Exclude utilities and financial firms (SIC codes 49 to 50 and 60 to 68, respectively); Exclude firms involved in spin-offs. Our finalsamples consist of firms that are present in TAQ, IBES, CRSP, and Compustat. We exclude firms with negative adverse selection costs (GH b 0) or adverse selectioncomponents greater than 1 (GH N 1). We obtain the number of analysts and the Earnings prediction error variable from IBES database. Profitability, market tobook ratio, and leverage ratio are calculated using data from Compustat.

Year CRSP and TAQ data GH N 0 and GH b 1 IBES data Compustat data

1993 658 630 545 5501994 730 704 595 6111995 788 767 605 6361996 815 787 625 6651997 834 810 643 6801998 863 849 673 6941999 867 862 657 6892000 831 828 632 6692001 829 828 649 6792002 880 878 706 7142003 928 919 760 7492004 926 926 786 7572005 969 955 816 7762006 992 974 836 7922007 991 881 862 8102008 1051 1022 887 827Total 13,952 13,620 11,277 11,298

86 R. Ravi, Y. Hong / Journal of Empirical Finance 25 (2014) 83–94

market dealers), which can cause quoted spread to be artificially inflated. Since reliable filtering of auto-quotes in TAQ is notpossible, only the BBO (best bid or offer)-eligible primary market (NYSE) quotes are used. The first trade of each day is discardedto avoid the effects of the opening procedure.

We estimate Eq. (1) once each year for each stock. Thus, we obtain one GH estimate for each firm in the sample, in each year ofthe sample period. A year is defined as the period from January 1st to December 31st. This also corresponds to the fiscal years ofthe sample firms. GH is the computed adverse selection component of the spread. Approximately 36% of the quoted spread for thetypical sample firm is due to the information asymmetry problem that the uninformed market maker faces due to the presence ofinformed traders in the market. DVAL_GH is the dollar value transformation of GH, which is obtained using the relation:

10 Somand Lun

DVALXGH ¼ GH.

Average trading price

� �� Average spreadð Þ: ð3Þ

The mean adverse selection cost of trading $100 in the basket of sample firms is about 14 cents while the median is about8 cents. In other words, the transaction costs increase on average by 14 cents (for every $100 worth of trade) due to informationasymmetry between investors (Table 2).

2.1.2. Proxies for firm-to-investor-IABy its nature, the magnitude of firm-to-investor-IA cannot be directly observed. In the existing literature, various proxies have

been devised for measuring this type of information asymmetry. While some of these proxies are based directly on firm anddisclosure characteristics (hereafter referred to as direct measures), others attempt to infer firm-to-market informationasymmetry from the precision of analysts' earnings forecasts (hereafter referred to as indirect measures). Some of the directmeasures include AIMR disclosure scores, S&P transparency and disclosure scores, and the magnitude of discretionary accruals.The indirect measures include the number of analysts' providing earnings forecasts for the firm, and the errors in analysts'forecasts. In this study, we are attempting to explore the relationship between firm to investor information asymmetry and theinformation asymmetry among investors. To the extent that analysts rely on their understanding of the firm to generate earningsforecasts, we propose that the analysts' based measures should be a good proxy for firm-to-investor-IA. Our hypothesis ofunimodal relationship should hold true not only for the direct measures but also for the analyst-based indirect measures of thefirm-to-investor-IA.

The Association for Investment and Management Research (AIMR) has published the annual rankings of corporate disclosurepractices for all years between 1982 and 1996. These scores have been widely used in academic research as an empirical proxy fordisclosure quality.10 According to the AIMR, these scores measure a firm's effectiveness in communicating with investors, and theextent to which its aggregate disclosure ensures that investors have the information necessary to make an informed judgment.

e of the studies that have used the AIMR disclosure scores include Botosan and Plumlee (2002), Gelb and Zarowin (2002), Bushee and Noe (2000), Langdholm (1996), and Lundholm and Myers (2002). Detailed description of the data can be found in Bushee and Noe (2000).

Table 2Descriptive statistics.Descriptive statistics are calculated across the complete sample period (1993 through 2008). GH is the fraction of the spread, which can be attributed to adverseselection cost (Glosten and Harris, 1988). DVAL_GH (=GH/average trading price × average spread) is the dollar-value transaction cost incurred due tointer-investor-IA. NAnal is the number of analysts following the stock (providing earnings estimates). EPE is the absolute difference between the average earningsforecast and the true earnings, expressed as a percentage of the share price. Abs_DA is the absolute value of the discretionary accruals, as estimated using themodified Jones model (Kothari et al., 2005). Market to book ratio (MB) is defined as (market value of equity + total debt) / total assets. Leverage (Lev) is definedas total debt divided by total assets. Ln_Tsize is the natural logarithm of the average daily transaction size. Ln_Treq is the natural logarithm of the average dailynumber of trades. Ln_Size is the market capitalization of the sample firm, while price represents the average daily trade price of the stock.

Panel A: Information asymmetry measures

GH DVAL_GH (Cents) NAnal EPE Abs_DA

Mean 0.360 0.14 8.50 0.041 0.08Median 0.365 0.08 9.03 0.002 0.04Standard deviation 0.151 0.17 2.16 0.124 0.18Maximum 0.986 4.07 44 0.393 3.93Minimum 0.007 0 1 0.001 0Percentile 99 0.702 0.74 34.12 0.0003 0.79Percentile 01 0.034 0.01 1.99 0.006 0

Panel B: Control variables

MB Lev Ln_Tsize Ln_Tfreq Ln_Size

Mean 3.263 0.566 6.81 3.49 13.97Median 2.045 0.593 6.86 3.49 13.90Standard deviation 15.918 0.342 0.67 1.73 1.68Maximum 828.055 22.787 9.45 8.54 20.04Minimum −326.855 0 4.94 −3.66 7.46Percentile 99 24.423 1.083 8.13 7.03 18.21Percentile 01 −4.531 0 5.43 −0.32 10.55

87R. Ravi, Y. Hong / Journal of Empirical Finance 25 (2014) 83–94

The non-availability of this data post 1996 makes them relatively unsuitable for our study. However, we do analyze a subset of oursample spanning from 1993 through 1996 using these scores. The results obtained are in concurrence with the ones reported inthis study using other proxies of firm-to-investor-IA.

Bhattacharya et al. (2013) use accruals as a proxy for poor earnings quality. They argue that the extent to which a firm'searnings (accruals) relate to its cash flows is related to innate factors such as the firm's business model and the operatingenvironment, and discretionary factors such as the reporting choices made by the mangers. They find that extreme positive andextreme negative discretionary accruals increase firm-to-investor information asymmetry. Following them, we use discretionaryaccruals as a measure of firm-to-investor-IA. To derive this measure of information asymmetry, we estimate the followingmodified Jones model of Kothari et al. (2005):

11 To r12 Brobetter c

ACCi; j;t

Assetsi; j;t−1¼ β0; j þ β1; j

1Assetsi; j;t−1

þ β2; jΔ Salesð Þi; j;tAssetsi; j;t−1

þ β3; jPPEi; j;t

Assetsi; j;t−1þ εi; j;t ð4Þ

ACC is total accruals (difference between earnings and cash flows); PPE is property plant and equipment; ΔSales is change

wherein sales relative to the previous year; i indexes firm while j indexes industries (defined by the two digit SIC code); and t indexesyear.11 The residual εi,j,t is taken to be the discretionary accrual.

Fried and Givoly (1982), O'Brien (1988), and Brown (1996) show that, financial analysts' earnings forecasts are goodmeasures of the market's expectations. Motivated by these studies, Dempsey (1989), Lobo and Mahmoud (1989), Brennan andSubramanyam (1995) and Coller and Yohn (1997) interpret the number of analysts forecasting the firm's earnings as a measureof firm-to-investor information asymmetry. According to these studies, the existence of analyst coverage should reducefirm-to-investor-IA. Hong et al. (2000) show that holding all else equal, the more analysts covering a company, the morefirm-specific information will be produced and the faster that information will be transmitted. The theoretical support for thismeasure can be traced to Blackwell and Dubins (1962). They show that opinions tend to converge as the amount of informationavailable about an unknown entity increases. Thus, ceteris paribus, the number of analysts covering a firm should be inverselyrelated to the magnitude of firm-to-investor-IA. As a related measure, Elton et al. (1984) and Best and Zhang (1993) usefinancial analysts' earnings forecast errors as a measure of firm-to-investor information asymmetry. According to their studies,higher errors in analysts' earnings forecasts should be related to larger firm-to-investor-IA. Analysts' earnings prediction error(mean forecast − current year earnings) is usually measured as a percentage of the stock price.12 All analyst data are obtainedfrom the Institutional Broker Estimate System (I/B/E/S). Table 2 Panel A presents some descriptive statistics corresponding tothe various measures of firm-to-investor-IA and inter-investor-IA used in this study.

educe the effect of outliers, all the variables in this paper are winsorized at 0.2 and 99.8 percentiles every year.us (1992), Christie (1987), and Pound (1988) discuss the merits of normalizing by stock price per share, and suggest that normalizing by price results in aharacterization of the importance of the error rather than normalizing by either mean forecasted earnings or actual earnings.

88 R. Ravi, Y. Hong / Journal of Empirical Finance 25 (2014) 83–94

2.2. Methodology

2.2.1. Cross-sectional regression testsAs a starting point of our analysis, we choose a simple second order polynomial regression13:

13 Weand to t14 Am(1991),15 Two

DVALXGH ¼ α þ β1 � IAþ β2 � IA2 þ other controlsð Þ þ ε ð5Þ

, DVAL_GH is as described in Eq. (3) (dollar-value adverse selection cost). We expect β1 N 0 and β2 b 0.

whereControl variables introduced in Eq. (5) are leverage ratio (Lev), market-to-book ratio (MB), average trading size (Tsize),

average number of trades (Tfreq), fiscal year profitability (Profit), return volatility (Ret_Vol) and average firm size. Boot andThakor (1993) demonstrate that the incentive for private information acquisition is positively related to financial leverage. This isbecause increased debt is associated with a greater probability of financial distress. To the extent that this creates valuationuncertainties, greater leverage could be positively associated with firm-to-investor-IA. However, as modeled by Ross (1977),greater leverage can signal the quality of a firm, and thus reduce uncertainty. Market-to-book ratio measures the growthopportunity of a firm relative to the value of the assets in place. This ratio should be related to information asymmetry betweenthe firm management and the outside investor (firm-to-investor-IA). However, as pointed out by Gaver and Gaver (1993), lowmarket-to-book ratio could indicate either lower growth opportunities or it could indicate financial distress. While the formerwould suggest potentially lower firm-to-investor-IA, the latter would lead to high firm-to-investor-IA.

Verrecchia (1983) suggests that managers are likely to provide more informative disclosures when they have good newsrather than bad news. This should lead to a relative decrease in firm-to-investor-IA in profitable years and a relative increase infirm-to-investor-IA during loss years. However, during a loss year, the manager may seize the opportunity to take a “big bath” andreveal all previously undisclosed bad news at once (Hutton et al., 2000). This would lead to a reduction in firm-to-investor-IAduring the bad years. Thus, the association between profitability and information asymmetry remains unclear. To the extent thatit might be affecting firm-to-investor-IA, we control for its effects in this study. Profitability is defined as the ratio of earningsbefore depreciation and amortization divided by book value of assets.

Welker (1995) suggests a non-linear relation between spread and price. To the extent that the GH measure is derived fromthe spread, this relation could potentially contaminate the results we are exploring in Eq. (5). In order to avoid this, wecontrol for the average daily trading price by introducing natural logarithm of firm size as a control variable in Eq. (5),(ln_Size = ln(Price) + ln(shares outstanding)). Existing literature presents firm size as one of the oldest proxies for measuringfirm-to-investor-IA.14 These studies suggest that larger firms have more publicly available information about future prospectsand therefore should be more transparent. However, to the extent that firm size is also related to various other firm features(performance, liquidity, etc.), we have introduced it as a control variable in our analysis. While this could potentially weakenthe explanatory power of “number of analysts” proxy (since it is highly correlated with size), the remaining proxies offirm-to-investor-IA(AIMR scores, error in analyst forecast, discretionary accruals) have a very low correlation with size andshould remain largely unaffected.15 The adverse selection cost of trading is also affected by various market microstructurefactors. We attempt to control for these effects by introducing daily number of trades (Tfreq), average daily trade size (Tsize)and return volatility (Ret_Vol) as control variables. Table 2, Panel B presents some descriptive statistics for the controlvariables. Table 3 presents the correlations across them.

The GH and DVAL_GH measures are negatively correlated with the number of analysts (Ln_Anal), size (Ln_Size), and themarket-to-book ratio (MB) at conventional levels. The changes of signs appearing on trading size (Ln_Tsize) and averageprice (Ln_price) between GH and DVAL_GH variables may be partially explained, as DVAL_GH is divided by average pricewhich affects the two variables. Correlations among independent variables do not seem to be large enough to result inmulticollinearity.

2.2.2. Piecewise linear regression testsWhile the polynomial regression could provide evidence in support of or against our hypothesis, the fitted functional form is

not necessarily accurate. Piecewise linear regression is an alternate approach for testing our hypothesis. In this case, apart fromtesting our hypothesis, the spline function can provide information about at what levels of information asymmetry the relationincreases versus declines.

We estimate piecewise linear regressions allowing for four changes in the slope coefficient on firm-to-investor-IA. Wepartition the sample firms into quintiles based on their opacity level, as measured by the various proxies for firm-to-investor-IA.We select four knots for the spline model, defined as:

qk ¼ Min Quintilekþ1� �

k ¼ 1;2;3;4 ð6Þ

do not include additional terms such as square root, cubic, quartic, etc. in model (4) because this exercise is simply to demonstrate a curvilinear relationhat end, we only require the existence of the second order derivative for the specified function. A simple quadratic polynomial is therefore sufficient.ong the studies that have used firm size to proxy firm-to-investor-IA include Vermaelen (1981), Atiase (1985), Bamber (1987), Diamond and VerrecchiaLlorente et al. (2002), and Chae (2005).firms of the same size can still have different levels of firm-to-investor information asymmetry.

where

Table 3Correlation among variables.The cross-sectional correlation is estimated in each year and the table below presents the average correlation across years. Profit is the earnings beforedepreciation and amortization divided by book value of assets.Ret_Vol is the idiosyncratic return volatility, estimated as the standard deviation of the market model residual. The rest of the variables are as defined in Table 2.

GH DVAL_GH Profit Anal Abs_DA EPE Ln_Tsize Ln_Tfreq Ln_Size Ret_Vol Lev MB

GH 1DVAL_GH 0.348⁎⁎ 1Profit 0 −0.089⁎⁎ 1Ln_Anal −0.434⁎⁎ −0.549⁎⁎ 0.132⁎⁎ 1Abs_DA 0.045⁎ −0.006 0.019 −0.036 1EPE 0.044⁎ 0.021 0.014 0.006 −0.023 1Ln_Tsize −0.539⁎⁎ 0.081⁎⁎ 0.006 0.350⁎⁎ −0.013 −0.038 1Ln_Tfreq −0.137⁎⁎ −0.658⁎⁎ 0.015 0.504⁎⁎ 0.051⁎ −0.004 −0.196⁎⁎ 1Ln_Size −0.242⁎⁎ −0.665⁎⁎ 0.204⁎⁎ 0.710⁎⁎ −0.01 0.009 0.078⁎⁎ 0.751⁎⁎ 1Ret_Vol −0.072⁎⁎ 0.311⁎⁎ −0.165⁎⁎ −0.196⁎⁎ 0.119⁎⁎ −0.04 0.260⁎⁎ 0.003 −0.325⁎⁎ 1Lev −0.021 −0.028 −0.186⁎⁎ 0.01 −0.037 −0.01 0.055⁎ 0.120⁎⁎ 0.102⁎⁎ −0.003 1MB −0.064⁎⁎ −0.113⁎⁎ 0.207⁎⁎ 0.166⁎⁎ 0.038 −0.042 0.037 0.155⁎⁎ 0.242⁎⁎ −0.018 −0.066⁎⁎ 1

*, **, and *** indicate significance at the 1%, 5%, and 10% levels, respectively.

16 AIMsubsam

89R. Ravi, Y. Hong / Journal of Empirical Finance 25 (2014) 83–94

Min(Quintilek + 1) is the minimum of the (k + 1) quintile (for a given measure of firm-to-investor-IA). The spline model is

wheregiven by:

y ¼ α þ β1Q1 þ β2Q2 þ β3Q3 þ β4Q4 þ β5Q5 þ control variablesþ ε ð7Þ

Q1;i ¼ xi if xi ≤ q1q1 otherwise Qk;i

k¼2;3;4

¼0 if xibqk−1xi−qk−1 if qk−1 ≤ xi ≤ qkqk otherwise

Q5;i ¼(0 if xibq4xi−q4 otherwise :

8<:

8<:

AIMR scores, Ln_Anal (natural logarithm of the number of analysts following the firm), EPE (analysts' earnings prediction error,scaled by stock price), and absolute value of DA (discretionary accruals) are the four measures of firm-to-investor informationasymmetry used in this analysis. Our primary interest here is in the five slope coefficients βi. For the hypothesized unimodalrelation to exist, we expect a series of positive coefficients followed by remaining negative coefficients.

3. Results

3.1. Cross-sectional regression analysis with linear specification

A large body of extant empirical literature suggests that firm-to-investor-IA and inter-investor-IA are linearly positivelyrelated. This view is intuitively supported by a commonly held view which suggests that information asymmetry results fromsome traders being privy to information prior to it being publicly available. Hasbrouck (1991) states that most useful privateinformation is essentially advance knowledge of public information. We use this as the starting point of our analysis.

Table 4 presents the results of estimating the following linear regression specification:

DVALXGH ¼ α þ β � IAþ γ other controlsð Þ þ ε ð8Þ

, DVAL_GH is the dollar value adverse selection cost of trading. IA are the various proxies for firm-to-investor-IA. The set of

wherecontrol variables includes leverage ratio, market-to-book ratio, average per day trade size, average daily trading frequency, returnvolatility and firm size.

We present results based on two disclosure quality measure of firm-to-investor-IA (AIMR scores and discretionary accruals) andtwo analyst based measures (number of analysts and error in analysts' predictions).16 In concurrence with existing literature, we findthat as themagnitude of discretionary accruals increases (in absolute value terms), the adverse selection cost of trading also increases.This suggests that an increase in firm-to-investor-IA leads to an increase in the inter-investor-IA, and vice versa.We find a similar resultusing error in earnings forecasts, whereby, increase in firm-to-investor-IA is likely tomake the task of forecasting future earningsmoredifficult and thus prone to greater error. We find that as the earnings prediction error increases, the inter-investor-IA too increases.

The number of analysts' providing earnings forecasts is a negative measure of firm-to-investor-IA. As the number of analystsfollowing the firm increases, more information about the firm is produced and therefore, the firm-to-investor-IA should decline.

R scores are available only until 1996. Therefore all analyses using this variable as a measure of firm-to-investor information asymmetry are based on aple of our data ranging from 1993 through 1996.

Table 4Robust linear regression with and without any controls.The dependent variable, DVAL_GH (GH/average trading price) ∗ (average spread) is the dollar value transaction cost incurred due to inter-investor informationasymmetry. Inter-investor information asymmetry GH is the fraction of the spread, which can be attributed to adverse selection cost (Glosten and Harris, 1988).Ln_Anal is the natural logarithm of the number of analysts following the stock (providing earnings estimates). EPE is the Absolute difference between the averageearnings forecast and the true earnings, expressed as a percentage of the share price. Abs(DA) is the absolute value of the discretionary accruals, as estimatedusing the modified Jones model (Dechow et al., 1995). Profit is earnings before depreciation and amortization divided by book value of assets. Market to book ratio(MB) is defined as (market value of equity + total debt) / total assets. Leverage (Lev) is defined as total debt divided by total assets. Ln_Tsize is the naturallogarithm of the average transaction size. Ln_Tfreq is the natural logarithm of the average daily number of trades. Ln_Size is the market capitalization of the samplefirm. Return volatility Ret_Vol is estimated as the standard deviation of the market model residual. AIMR represents the disclosure scores as published by theAssociation for Investment and Management Research (AIMR).We use firm-clustered bootstrap standard errors (Cameron et al., 2008) in order to draw inferences from the following regression. The standard errors are inparentheses. ⁎⁎⁎ implies significance at 1%, ⁎⁎ implies significance at 5%, and ⁎ implies significance at 10%.

Dependent variable (DVAL_GH)

AIMR −0.013⁎⁎

(0.005)−0.006⁎

(0.003)Abs_DA 0.073⁎⁎⁎

(0.013)0.011⁎⁎⁎

(0.001)EPE 0.050⁎⁎⁎

(0.002)0.027⁎⁎⁎

(0.002)Ln_Anal −0.076⁎⁎⁎

(0.002)−0.040⁎⁎⁎

(0.002)Lev 0.001⁎⁎ 0.011⁎⁎⁎ 0.006 0.009⁎⁎

MB 0.012⁎ 0.00009 0.00007 0.00003Ln_Tsize −0.022⁎ −0.008⁎⁎⁎ −0.003⁎ 0.015⁎⁎⁎

Ln_Tfreq −0.051⁎⁎ −0.057⁎⁎⁎ −0.047⁎⁎⁎ −0.042⁎⁎⁎

Profit −0.001 −0.004 −0.005 −0.008⁎⁎

Ret_Vol 1.472⁎⁎⁎ 3.289⁎⁎⁎ 2.382⁎⁎⁎ 2.267⁎⁎⁎

Ln_Size −0.064⁎⁎⁎ −0.012⁎⁎⁎ −0.010⁎⁎⁎ −0.003⁎⁎

Adj. R2 0.01 0.479 0.001 0.506 0.054 0.574 0.197 0.578

90 R. Ravi, Y. Hong / Journal of Empirical Finance 25 (2014) 83–94

We find a negative relation between the number of analysts and the inter-investor-IA, suggesting that as the number of analystsincreases, the inter-investor-IA is likely to decline.

3.2. Exploring non-linearity

While the linear relationship between firm-to-investor and inter-investor information asymmetry, as observed in Section 3.1,is true on average, this paper argues that this could potentially be a misrepresentation of the true relationship. Building on thetheoretical arguments of Kim and Verrecchia (1991), we propose that the nature of the relationship should be unimodal. Thissubsection attempts to explore this nonlinearity using two methods, first a simple polynomial regression with quadraticspecification, and second a spline regression.

3.2.1. Cross-sectional regression analysis with quadratic specificationWe use a simple polynomial regression (Eq. (5)), to explore the unimodal relation between firm-to-investor-IA and the

inter-investor-IA. The model estimated in this subsection is of the form:

17 Wedispersibetweeactual eanalysts

DVALXGHt; j ¼ α þ β1 � IAt; j þ β2 � IA2t; j þ β3 � Levt; j þ β4 � ln MBt; j

� �þ β5 � ln Tsizet; j

� �β6 � ln Tfreqt; j

� �þ β7 � RetXVolt; j þ β8 � ln Sizet; j

� �þ β9 � Profitt; j þ γi;tD1993−2002 þ εt; j

ð9Þ

, DVAL_GHt,j measures the level of inter-investor-IA for firm j in year t, as measured by the per share dollar value adverse

whereselection cost of trading firm j's shares in year t.17 The adverse selection cost is expressed in dollars per $100 traded. IA is themeasure of the firm-to-investor-IA. According to the proposed unimodal relation discussed in Section 1, we expect GH to behighest for firms that are neither very opaque nor very transparent. Opaque firms represent firms with high firm-to-investor-IA,while transparent firms represent those with low firm-to-investor-IA. Therefore, in the notation of Eq. (9), we expect the estimateof β1 to be positive and the estimate of β2 to be negative (Table 5).

Eq. (9) is estimated as a pooled panel model with firm level clustering. The control variables remain as before and DUM is adummy variable which takes the value 1 if the firm incurred a loss in the given year, and zero otherwise. The subscripts t and jrepresent year t and firm j. Year dummies D are used to capture any possible time trend. Table 5 presents the estimated

thank an anonymous referee for suggesting “dispersion of analysts' forecast” as an alternate measure of between-investor information asymmetry. Theon of earnings forecast would signify the degree of disagreement among users of firm information and thus proxy for the level of information asymmetryn investors. (Note: This is different from the analysts' forecast error that we have used in the paper as a proxy for firm-investor IA. The error betweenarnings and forecast earnings can signify firm-to-investor IA.) We replicate our analysis using dispersion of analysts' forecast (coefficient of variation in' earnings forecast) instead of DVAL_GHt,j. The results remain qualitatively unchanged.

Table 5Quadratic regression analysis.We regress DVAL_GH on linear and second order terms of the various proxies for firm-to-investor-IA (as defined in Table 4). The coefficients are estimated as apooled panel model. The significance is based on the H0: Average coefficient is equal to zero. Inferences are based on firm-clustered bootstrap standard errors(Cameron et al., 2008). The standard errors are in parentheses. ⁎⁎⁎ implies significance at 1%, ⁎⁎ implies significance at 5%, and ⁎ implies significance at 10%.

Expected sign Model 1a Model 2 Model 3 Model 4

AIMR + 0.0003⁎⁎

(0.0001)AIMR2 − −0.002⁎

(0.0009)Ln_Anal + 0.037⁎⁎⁎

(0.006)Ln_Anal2 − −0.023⁎⁎⁎

(0.002)EPE + 0.413⁎⁎⁎

(0.055)EPE2 − −4.160⁎⁎⁎

(0.927)Abs_DA + 0.003⁎⁎⁎

(0.001)Abs_DA2 − −0.001⁎⁎

(0.0004)Lev 0.118⁎⁎⁎ 0.011⁎⁎⁎ 0.003⁎⁎ 0.011⁎⁎⁎

MB 0.002 0.000 0.000 0.000Ln_Tsize 0.005⁎⁎⁎ 0.012⁎⁎⁎ −0.004⁎ −0.007⁎⁎⁎

Ln_Tfreq −0.044⁎⁎⁎ −0.042⁎⁎⁎ −0.047⁎⁎⁎ −0.057⁎⁎⁎

Profit −0.046⁎ −0.008⁎⁎ −0.005 −0.004Ret_Vol 2.173⁎⁎⁎ 2.245⁎⁎⁎ 2.294⁎⁎⁎ 3.295⁎⁎⁎

Ln_Size −0.014⁎⁎⁎ −0.006⁎⁎⁎ −0.009⁎⁎⁎ −0.012⁎⁎⁎

Adj. R2 0.614 0.585 0.565 0.517

a Model 1 is estimated over 1993 through 1996, limited by the availability of AIMR disclosure scores. The rest of the models have been estimated over thecomplete sample period (1993 through 2008).

91R. Ravi, Y. Hong / Journal of Empirical Finance 25 (2014) 83–94

coefficients. Consistent standard errors are estimated according to Cameron et al. (2008)'s specification. We find β1 estimate to bepositive and β2 estimate to be negative for all measures of firm-to-investor-IA (IA). The results suggest that as firm-to-investor-IAincreases (i.e., firms become more opaque or disclosure quality declines), GH increases (positive β1) because increasedfirm-to-investor-IA provides more opportunities for smart investors to derive greater private benefits from firm-specificinformation. However, simultaneously increasing search costs eliminate the marginally informed investors. This increases theproportion of uninformed traders in the market, which subsequently causes GH to decline (negative β2).

3.2.2. Piecewise linear (spline) regressionIn this section,we fit piecewise linear regression to explore the relationship between firm-to-investor-IA and inter-investor-IA.We

use two analyst basedmeasures (error in analysts' forecasts, and the number of analysts following the firm) and twodisclosure relatedmeasure (AIMR scores, and, absolute value of discretionary accruals) for this analysis. Following the definitions laid out inSection 2.2.2, we construct a set of five new variables corresponding to each of the proxies for firm-to-investor-IA. Table (6) presentsthe results of this analysis. Columns 2, 4, 6, and8 present the spline coefficients from the base case regression (no controls). Columns 3,5, 7, and9 present the splinemodel (Eq. (7)) coefficients, controlling for leverage ratio,market to book ratio, average daily trading size,average trading frequency, return volatility, firm size and profitability of the firm.18 AIMR scores and number of analyst' measures areinverse measures of firm opacity. Thus, as the AIMR score or the number of analysts increases, the firm-to-investor-IA declines (thefirm becomes more transparent). The opposite is true for the remaining twomeasures. As the earnings forecasts errors increase, or asthe discretionary accruals increase (in absolute value terms), the firm-to-investor-IA increases (the firm becomes less transparent).

The observed relationship is consistent with the hypothesized unimodal relationship between the two types of informationasymmetries. As AIMR scores increase (firm becomes more transparent), the inter-investor information asymmetry increasesover the first quintile and declines across the remaining four quintiles. Similarly, as discretionary accruals increase (in absoluteterms, Abs_DA), the firm becomes more opaque. We find that the inter-investor-IA increases over the first four quintiles of Abs_DAand declines across the fifth quintile. Earnings prediction error (EPE) provides similar results, whereby firm-to-investor andinter-investor-IAs are positively related across the first four quintiles and negatively related over the fifth quintile. The number ofanalysts' variable while displaying the hypothesized unimodal relationship in the univariate analysis seems to break down across

18 Welker (1995) suggests a non-linear relation between spread and price. We run two alternate specifications of Eq. (7), first with an additional quadratic priceterm included as a control variable, and second with a spline (with one node) for the price. Our results remain unchanged. For the sake of brevity, thesespecifications are not reported in Table 6.

Table 6Piecewise linear ordinary least squares regression.Piecewise linear ordinary least squares regression of DVAL_GH on various proxies of firm-to-investor information asymmetry, and other firm characteristics forour complete sample (described in Table 1).The data is split into quintiles (by year) based on the absolute value of Discretionary Accruals (DA).DA_Q1 = abs(DA) if abs(DA) is within the first abs(DA) quintile in a given year. If abs(DA) is in second through 5th quintile, then DA_Q1 = Min value of abs(DA)in the second quintile.DA_Q2 = 0 if abs(DA) is within the first quintile. If abs(DA) is within the second quintile, then DA_Q2 = abs(DA) − (Min value of abs(DA) in the secondquintile). If abs(DA) is within third through 5th quintile, DA_Q2 = Min value of abs(DA) in the third quintile.DA_Q3 = 0 if abs(DA) is within the first or the second quintile. If abs(DA) is within the third quintile, then DA_Q3 = abs(DA) − (Min value of abs(DA) in thethird quintile). If abs(DA) is within fourth through fifth quintile, DA_Q3 = Min value of abs(DA) in the fourth quintile.DA_Q4 = 0 if abs(DA) is within the first, second or the third quintile. If abs(DA) is within the fourth quintile, then DA_Q4 = abs(DA) − (Min value of abs(DA) inthe fourth quintile). If abs(DA) is within fifth quintile, DA_Q4 = Min value of abs(DA) in the fifth quintile.DA_Q5 = 0 if abs(DA) is within the first, second, third or the fourth quintile. Otherwise, DA_Q5 = abs(DA) − (Min value of abs(DA) in the fifth quintile).Similar variables are constructed for Earnings prediction errors (EPE_Q1 through Q5), for number of analysts (Anal_Q1 through Q5), and AIMR disclosure scores.(The AIMR models are estimated over the subsample period ranging from 1993 through 1996 because the Association for Investment and Management Research(AIMR) published the annual rankings of corporate disclosure practices for all years between 1982 and 1996.)The set of control variables consists of Market-to-book ratio (MB) defined as (market value of equity + total debt) / total assets, Leverage (Lev) is defined as totaldebt divided by total assets. Ln_Tsize is the natural logarithm of the average transaction size in the given year. Ln_Tfreq is the natural logarithm of the average dailynumber of trades. Ln_Size is the market capitalization of the sample firm calculated as the average of the daily market capitalization (closing price ∗ sharesoutstanding). Return volatility Ret_Vol is estimated as the standard deviation of the market model residual. Profit is earnings before depreciation and amortizationdivided by book value of assets.The significance levels are based on consistent standard errors calculated according to Cameron et al. (2008).The fitted model is given by Eq. (7): DVAL_GH = α0 + α1Q1 + α2Q2 + α3Q3 + α4Q4 + α5Q5 + control variables + ε where Q1 through Q5 represent thevariables constructed as defined above.

Dependent variable (DVAL_GH)

AIMR_Q1 0.037⁎ 0.009⁎

AIMR_Q2 −0.006⁎ −0.002⁎

AIMR_Q3 −0.195⁎⁎⁎ −0.002⁎⁎

AIMR_Q4 −0.060⁎ −0.001⁎

AIMR_Q5 −0.067⁎⁎ −0.020⁎⁎

DA_Q1 0.07⁎ 0.112⁎⁎

DA_Q2 2.147⁎⁎ 0.170⁎⁎

DA_Q3 1.016⁎⁎ 0.062⁎⁎⁎

DA_Q4 −0.017⁎ 0.051⁎

DA_Q5 −0.056⁎ −0.020⁎⁎⁎

EPE_Q1 6.553⁎⁎⁎ 0.1681⁎

EPE_Q2 4.131⁎⁎⁎ 10.382⁎⁎⁎

EPE_Q3 0.470⁎⁎⁎ 12.588⁎⁎

EPE_Q4 0.336⁎⁎ 2.988⁎

EPE_Q5 −0.141⁎⁎ −0.233⁎

Anal_Q1 0.110⁎ 0.035⁎⁎⁎

Anal_Q2 0.070⁎⁎ −0.016⁎⁎⁎

Anal_Q3 −0.028⁎⁎ −0.004⁎⁎⁎

Anal_Q4 −0.029⁎ 0.011⁎

Anal_Q5 −0.701⁎ −0.028⁎⁎⁎

Lev 0.005⁎ .044⁎⁎⁎ .003⁎⁎⁎ .007⁎

MB 0.012⁎⁎ .001 4.359E−005 2.697E−005Ln_Tsize −0.026⁎⁎ − .038⁎⁎⁎ − .012⁎⁎⁎ .014⁎⁎⁎

Ln_Tfreq −0.074⁎⁎⁎ − .057⁎⁎⁎ − .048⁎⁎⁎ − .041⁎⁎⁎

Profit −0.002 −0.002 −0.003 −0.006⁎

Ret_Vol 1.285⁎⁎ 3.869⁎⁎⁎ 2.237 2.067⁎⁎⁎

Ln_Size −0.046⁎⁎⁎ − .045⁎⁎⁎ − .025⁎⁎⁎ − .029⁎⁎⁎

Adj. R-Square 0.023 0.486 0.018 0.545 0.098 0.548 0.213 0.562

*, **, and *** indicate significance at the 1%, 5%, and 10% levels, respectively.

92 R. Ravi, Y. Hong / Journal of Empirical Finance 25 (2014) 83–94

the fourth quintile in the controlled model. A possible explanation for this could be that the number of analysts' proxy might be aless accurate measure of firm-to-investor-IA, as it is highly correlated with firm size.19

Fig. 1A, and B presents the fitted spline models corresponding to the results reported in columns 3, and 5 of Table 6. Thesefigures represent the estimated relationship between firm-to-investor-IA and inter-investor-IA, obtained using the regressionmodel with the control variables.

Overall, the results suggest that for most firms, as firm-to-investor-IA declines, so does inter-investor-IA. However, thisrelationship flips for the most opaque firms. For twenty percent of the most opaque firms, as the firm-to-investor-IA declines, theinter-investor-IA increases.20 For these firms, greater quantities of and/or better quality information released by the firm seems to

19 A caveat is warranted here, whereby; the unexpected behavior of the “Number of analysts” proxy could also be potentially alluding to a yet anotherunexplored dimension in the relationship between the two types of information asymmetries.20 Here twenty percent figure emerges by design since we split the firms into quintile groups. The exact percentage of firms for which the two types ofinformation asymmetries are negatively related is an empirical issue.

0.65

0.655

0.66

0.665

0.67

0.675

0.68

2 2.5 3 3.5 4 4.5 5

DV

AL

_GH

ln(AIMR Scores)

0.54

0.542

0.544

0.546

0.548

0.55

0.552

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16

DV

AL

_GH

Abs(Discretionary Accruals)

A

B

Fig. 1. Spline regressions of inter-investor-IA on firm-to-investor-IA measures. A: AIMR scores (natural logarithm) × axis denotes the disclosure quality rank asstated in AIMR 1993 to 1996 report. Higher rank implies better quality disclosure thereby more transparent firm. AIMR study ranked firms on scale of 1 to 100.Each segment in the above plot represents the best linear fit for successive firm quintiles (created on the basis of their respective AIMR score). The fitted equationis: DVAL_GH = 0.646 + 0.009 × Q1 − 0.002 × Q2 − 0.002 × Q3 − 0.001 × Q4 − 0.002 × Q5. The independent variables Q1 through Q5 are created asdescribed in Eq. (7). The coefficients are as reported in column 3 of Table 6. B: Discretionary accruals. We use the modified Jones model (Dechow et al., 1995)to estimate the discretionary accruals in the firm. Absolute value of the estimated discretionary accruals (ADA) is used as a proxy for firm-to-investor informationasymmetry. Greater ADA thus suggests relatively more opaque firm. The x-axis denotes the quintiles constructed on the basis of the absolute value of thediscretionary accruals. The fitted equation is: DVAL_GH = 0.541 + 0.112 × Q1 + 0.17Q2 + 0.062 × Q3 + 0.051 × Q4 − 0.02 × Q5. The independent variablesQ1 through Q5 are created as described in Eq. (7). The coefficients are as reported in column 5 of Table 6.

93R. Ravi, Y. Hong / Journal of Empirical Finance 25 (2014) 83–94

provide more material to investors who are better at processing public signals in order to create private benefits and vice versa.Thus, a decline in firm-to-investor-IA leads to an increase in inter-investor-IA. The result is intuitively appealing in thatdisclosures from opaque firms hold greater private information potential for investors who are better and faster at processing it.As the firm becomes increasingly transparent, this advantage declines.

4. Conclusion

This paper examines the relation between opacity of a firm (firm-to-investor-IA) and the information asymmetry betweeninvestors (inter-investor-IA). We provide evidence showing that the relation between firm-to-investor-IA and inter-investor-IA isunimodal. This result holds even after controlling for various market microstructure concerns, including level of debt, and otherfirm characteristics such as size andmarket-to-book ratios. We find that for the majority of firms (80% based on our spline design)as the firm-to-investor-IA declines, so does the inter-investor-IA. However for some of the most opaque firms (the remaining20%), the two types of information asymmetries are negatively related. This suggests that the benefits of a small improvement intheir disclosure quality are more likely to be captured by investors with superior processing ability.

Intuitively, the relation between the two types of information asymmetry is determined by the interplay of information searchcost and the benefits derived from trading with superior information. While the former is likely to dissuade informed trading,which would lead to a decline in inter-investor-IA, the latter is likely to encourage informed trading, which, in turn, would lead toan increase in inter-investor-IA. Our results suggest that as firm-to-investor-IA declines, that is, opaque firms become moretransparent, the benefits of superior information exceed the information search costs, leading to an increase in informed tradingand therefore an increase in inter-investor-IA. However, beyond a point, as the firm becomes increasingly more transparent, itbecomes simpler for a larger fraction of the market to quickly process the new disclosure. Thus, the cost of creating privatebenefits from this disclosure increases while the potential benefits declines consequently leading to a decline in inter-investor-IA.

References

Amihud, Y., Mendelson, H., 1988. Liquidity and asset prices—financial management implications. Financ. Manag. 17, 5–15.

94 R. Ravi, Y. Hong / Journal of Empirical Finance 25 (2014) 83–94

Atiase, R.K., 1985. Pre-disclosure information, firm capitalization, and security price behavior around earnings announcements. J. Account. Res. 23, 21–36.Bamber, L.S., 1987. Unexpected earnings, firm size, and trading volume around quarterly earnings announcements. Account. Rev. 62, 510–532.Bessembinder, H., 2003. Issues in assessing trade execution costs. J. Financ. Mark. 6, 233–257.Best, R.W., Zhang, H., 1993. Alternative information sources and the information content of bank loans. J. Financ. 48, 1507–1522.Bhattacharya, N., Desai, H., Venkataraman, K., 2013. Should earnings quality affect information asymmetry: evidence from trading costs. Contemp. Account. Rev.

30 (2), 482–516.Blackwell, D., Dubins, L., 1962. Merging of opinions with increasing information. Ann. Math. Stat. 33, 882–886.Boot, A.W.A., Thakor, A.V., 1993. Security design. J. Financ. 48, 1349–1378.Botosan, C.A., Plumlee, M.A., 2002. A re-examination of disclosure level and expected cost of equity capital. J. Account. Res. 40, 2–40.Brennan, M., Subramanyam, A., 1995. Investment analysis and price formation in security markets. J. Financ. Econ. 38, 361–381.Brous, P.A., 1992. Common stock offerings and earnings expectations: a test of the release of unfavorable information. J. Financ. 47, 1517–1536.Brown, L.D., 1996. Analyst forecasting errors and their implications for security analysis: an alternative perspective. Financ. Anal. J. 52, 40–47.Brown, S., Hillegeist, S., 2007. How disclosure quality affects the level of information asymmetry. Rev. Acc. Stud. 12, 443–477.Bushee, B.J., Noe, C.F., 2000. Corporate disclosure practices, institutional investors, and stock return volatility. J. Account. Res. 38, 171–202.Cameron, A.C., Gelbach, J.B., Miller, D.L., 2008. Bootstrap-based improvements for inference with clustered errors. Rev. Econ. Stat. 90, 414–427.Chae, J., 2005. Trading volume, information asymmetry, and timing information. J. Financ. 60, 413–442.Christie, A.A., 1987. On cross-sectional analysis in accounting research. J. Account. Econ. 9, 231–258.Coller, M., Yohn, T.L., 1997. Management forecasts and information asymmetry: an examination of bid–ask spreads. J. Account. Res. 35, 181–191.Dechow, P.M., Sloan, R.G., Sweeney, A.P., 1995. Detecting earnings management. Account. Rev. 70, 193–225.Dempsey, S.J., 1989. Predisclosure information search incentives, analyst following, and earnings announcement response. Account. Rev. 64, 748–757.Desaia, C.A., Savickas, R., 2010. On the causes of volatility effects of conglomerate breakups. J. Corp. Financ. 16 (4), 554–571.Diamond, D.W., 1985. Optimal release of information by firms. J. Financ. 40, 1071–1094.Diamond, D.W., Ver'recchia, R.E., 1991. Disclosure, liquidity, and the cost of capital. J. Financ. 46, 1325–1359.Elton, E.J., Gruber, M.J., Gultekin, M.N., 1984. Professional expectations: accuracy and diagnosis of errors. J. Financ. Quant. Anal. 19, 351–363.Fried, D., Givoly, D., 1982. Financial analysts' forecasts of earnings: a better surrogate for market expectations. J. Account. Econ. 4, 85–107.Gaver, J.J., Gaver, K.M., 1993. Additional evidence on the association between the investment opportunity set and corporate financing, dividend, and

compensation policies. J. Account. Econ. 16, 125–160.Gelb, D.S., Zarowin, P., 2002. Corporate disclosure policy and the informativeness of stock prices. Rev. Acc. Stud. 7, 33–52.Glosten, L.R., Harris, L.E., 1988. Estimating the components of the bid/ask spread. J. Financ. Econ. 21, 123–142.Hakansson, N.H., 1977. Interim disclosure and public forecasts—an economic analysis and a framework for choice. Account. Rev. 52, 396–426.Hasbrouck, J., 1991. Measuring the information content of stock trades. J. Financ. 46, 179–207.Healy, P.M., Palepu, K.G., 2001. Information asymmetry, corporate disclosure, and the capital markets: a review of the empirical disclosure literature. J. Account.

Econ. 31, 405–440.Healy, P.M., Hutton, A.P., Palepu, K.G., 1999. Stock performance and intermediation changes surrounding sustained increases in disclosure. Contemp. Account. Res.

16, 485–520.Heflin, F., Shaw, K.W., Wild, J.J., 2005. Disclosure quality and market liquidity: impact of depth quotes and order sizes. Contemp. Account. Res. 22, 829–865.Hong, H., Lim, T., Stein, J.C., 2000. Bad news travels slowly: size, analyst coverage, and the profitability of momentum strategies. J. Financ. 55, 265–295.Huson, M.R., MacKinnon, G., 2003. Corporate spinoffs and information asymmetry between investors. J. Corp. Financ. 9, 481–503.Hutton, A.P., Gregory, S.M., Douglas, J.S., 2000. Effective Voluntary Disclosure. Harvard Business School Working Paper. Harvard Business School.Kandel, E., Pearson, N.D., 2005. Differential interpretation of public signals and trade in speculative markets. J. Polit. Econ. 103, 831–872.Ke, B., Ramaligegowda, S., 2005. Do institutional investors exploit the post-earnings announcement drift? J. Account. Econ. 39, 25–53.Kim, O., Verrecchia, R.E., 1991. Market reaction to anticipated announcements. J. Financ. Econ. 30, 273–309.Kim, O., Verrecchia, R.E., 1994. Market liquidity and volume around earnings announcements. J. Account. Econ. 17, 41–67.Kothari, S.P., Leone, A., Wasley, C., 2005. Performance matched discretionary accruals measures. J. Account. Econ. 39, 163–197.Krinsky, I., Lee, J., 1996. Earnings announcement and the components of the bid–ask spread. J. Financ. 51, 1523–1535.Lang, M.H., Lundholm, R.J., 1996. Corporate disclosure policy and analyst behavior. Account. Rev. 71, 467–492.Lee, C.M.C., 1992. Earnings news and small traders. J. Account. Econ. 15, 265–302.Lee, C.M.C., Radhakrishna, B., 2000. Inferring investor behavior: evidence from TORQ data. J. Financ. Mark. 3, 83–111.Lee, C.M.C., Ready, M.J., 1991. Inferring trade direction from intraday data. J. Financ. 46, 733–746.Lee, C.M.C., Mucklow, B., Ready, M.J., 1993. Spreads, depths, and the impact of earnings information: an intraday analysis. Rev. Financ. Stud. 6, 345–374.Lin, J., Sanger, G., Booth, G., 1995. Trade size and components of the bid–ask spreads. Rev. Financ. Stud. 8, 1153–1183.Llorente, G.M., Saar, R.G., Wang, J., 2002. Dynamic volume-return relation of individual stocks. Rev. Financ. Stud. 15, 1005–1047.Lobo, G.J., Mahmoud, A.W.A., 1989. Relation between differential amounts of prior information and security return variability. J. Account. Res. 27, 116–134.Lundholm, R.J., 1991. Public signals and the equilibrium allocation of private information. J. Account. Res. 29, 322–349.Lundholm, R., Myers, L.A., 2002. Bringing the future forward: the effect of disclosure on the returns-earnings relation. J. Account. Res. 40, 809–839.Neal, R., Wheatley, S., 1998. Adverse selection and bid–ask spreads: evidence from closed-end funds. J. Financ. Mark. 1, 121–149.O'Brien, P.C., 1988. Analysts' forecasts as earnings expectations. J. Account. Econ. 10, 53–83.Odders-White, E.R., 2000. On the occurrence and consequences of inaccurate trade classification. J. Financ. Mark. 3, 259–286.Piwowar, M., Wei, L., 2006. The sensitivity of effective spread estimates to trade-quote matching algorithms. Electron. Mark. 16, 112–129.Pound, J., 1988. The information effects of takeover bids and resistance. J. Financ. Econ. 22, 207–227.Ross, S.A., 1977. The determination of financial structure: the incentive-signaling approach. Bell J. Econ. 8, 23–40.Vergote, O., 2005. How to match trades and quotes for NYSE stocks. Working Paper.Vermaelen, T., 1981. Common stock repurchases and market signaling: an empirical study. J. Financ. Econ. 9, 139–183.Verrecchia, R.E., 1983. Discretionary disclosure. J. Account. Econ. 5, 179–194.Welker, M., 1995. Disclosure policy, information asymmetry and liquidity in equity markets. Contemp. Account. Res. 11, 801–827.