Thinly Traded Growth Stocks: A Joint Examination of Transparency in Communication and the Trading Platform

Abstract

When thinly traded growth stocks (TTGS) listed on a secondary exchange experience difficulty in gaining investors' attention, one possible solution is to increase the intensity of disclosure. However, if the stock is traded on a quote-driven system, market makers can collude to maintain wide bid–ask spreads that discourage firms from disclosing. As a result, TTGS traded on a quote-driven system can face a liquidity trap that can prevent them from harvesting the benefits of increased disclosure activities. In this paper, we argue that the well-documented negative relation between disclosure and the bid–ask spread is likely to be moderated by the type of protocol chosen by exchanges to handle the trading of TTGS. To test our theory we use a unique setting created by the introduction of a hybrid order-driven protocol for TTGS in the UK. Following an increase in the disclosure activity by a TTGS, we find that the magnitudes of the predicted reductions in the bid–ask spreads are dependent on whether the TTGS switch their trading protocols.

Acknowledgements

We thank two anonymous reviewers and Laurence van Lent (the editor) for detailed suggestions on improving the paper. We also thank Richard Payne, Christian Leuz, Stephen Penman, Marco Trombetta, and conference participants at the 2010 seminar on Methodological and Empirical Advances in Financial Analysis, the University of Sydney, the 2010 European Accounting Association annual congress, and the 2010 EIASM Workshop on Accounting and Economics for helpful comments.

Notes

¹Significant literature exists on the modelling of how the transparency of the trading mechanism affects price formation – see, for instance, Chapter 10 of De Jong and Rindi (2009). That literature focuses upon the relative gains to various types of traders under various transparency settings and not the potential benefits to firms, which is our main concern here.

²See, for instance, Christie et al. (1994).

³Interestingly, Nimalendran and Petrella (2003) analyse the introduction of a hybrid (order protocol with market makers) market to Italy.

⁴See, for instance, Mendoza (2008).

⁵For example, AIM firms are not required to have had prior trading nor do they have to seek prior shareholder approval for transactions, there is no minimum market capitalization or minimum public float. The only disclosure obligation for firms is the ‘general duty of disclosure requiring information which the issuer reasonably considers necessary to enable investors to form a full understanding of the financial position of the applicant’. Considering that AIM firms predominantly target institutional investors with specialist knowledge, disclosure strategies of AIM firms tend to relate to voluntary dissemination of relevant information to those investors.

⁶In order to additionally check whether the firm's life cycle stage (age) might differentially affect the bid–ask spread for the switching subsample, we hand collect the data on all the firms' births (i.e. on the year of business inception). We find that the average life cycle (age) of the two subsamples is not statistically significantly different, that is, differences in age are not driving differences in spread.

⁷In the UK, firms are required to list information on a primary information provider like the RNS before talking to individual investors. The RNS is the primary timely source of information that newswire contributors use to base a report.

⁸See Bushee and Miller (2012).

⁹In order to test the validity of our disclosure measure, we also use the RNS release instead of all newswires to compute the variables WIRES. We are thankful to the anonymous referee for this suggestion.

¹⁰By ranking firms into quintiles of the variable WIRES separately for each subsample of firms (switchers and non-switchers) and pooling over all quarterly periods for that particular subsample, we are able to use firms as their own controls in the regression models estimated separately for switchers and non-switchers. In this way, we are able to analyse the effect of disclosure intensity on the bid–ask spread before and after the quarter in which a particular firm switches to SETSmm (in the regression of switchers), or before and after the quarter of the introduction of SETSmm (in the regression of non-switchers).

¹¹This type of calculation means that the number of high disclosers is always 20% over the whole sample period, but within a particular quarter it can differ.

¹²Defined in terms of all newswires.

¹³Variable WIRES measured by RNS exhibits similar behaviour.

¹⁴Note that the coefficient on SET is the partial derivative of SPREAD with respect to SET, holding DIS constant at zero and that DIS = 0 is outside the data range. Hence, in order to provide a more meaningful interpretation of the coefficient on SET, in the interaction models, we carry out the centring of variable DIS by subtracting the sample mean of DIS from DIS values in each observation, so that the mean of DIS is now zero. This way, the coefficient on SET now shows the difference in SPREAD between pre- and post-switching periods at the mean value of DIS. Also, note that the slope coefficients, their standard errors and t-test are the same in centred as in uncentred equations (Aiken and West, 1991).

¹⁵We thank the anonymous referee for this suggestion.

¹⁶Since the period of estimation is before the introduction of SETSmm, variables SET and DIS*SET take the value of zero by construction.

¹⁷We observe similar patterns in data for the RNS specification of variable WIRES.

¹⁸Datastream expresses daily trading volume in thousands of GBPs. Hence, the construct PRIMPACT captures the percentage by which a share price moves in £1000 of daily trading volume.

¹⁹In order to check whether the life cycle might be a correlated omitted variable, we include the age of the firm as an additional covariate (unreported) and find that it does not affect our results. However, we recognize that the age of the firm might not be an appropriate proxy for lifecycle as market conditions might differ for firms of a similar age but from different industries, so we collect information on the SIC industry classification for each firm and compare the industry compositions between pre- and post-switching periods. We find that the observations are fairly evenly spread among the industries. To further check the possibility that firms before and after the switch might be in a different life cycle, we collect the data on turnover growth (see the study by Anthony and Ramesh, 1992 on using turnover growth as a proxy for life cycle) for all companies over the key SETSmm adoption period from 2005 to 2007 and find statistically insignificant difference in mean growth rates between the pre- and post-switching periods. Moreover, in unreported analyses, we include three additional firm-level controls simultaneously in the model: (1) turnover growth, (2) its interaction with DIS which should capture the differential behaviour between low and high growth companies (i.e. between firms in early versus those in the later stages of life cycle) in terms of their disclosures' impact on SPREAD, and (3) in order to test whether potential life cycle differences between firms before and after their switch to SETSmm influence how their disclosure activities affect spread we add a three-way interaction between growth, DIS and SET. We find that our original results continue to hold and that the life cycle does not appear to influence how disclosure affects spread either pre- or post-switching to SETSmm. We confirm the conjecture that the type of trading is the main mechanism through which TTGS seem to reinforce the effects of their disclosure activities on bid–ask spread.

²⁰In order to address a potential concern that disclosure levels are significantly different before compared to after the introduction of SETSmm (i.e. it is theoretically possible that majority of low (high) disclosers are concentrated in periods before (after) the switch to SETSmm), which might bias the results from regression (2), we construct a balanced sample where the disclosure distributions are comparable before and after the switch. Within each disclosure rank (1–5) and across the two periods (pre- and post-SETSmm) we match observations by size so that the disclosure quintile composition is constant across the two periods. Using the matched sample we re-estimate model (2) and find that our original findings continue to hold and that they do not appear to be driven by differences in disclosure distributions before and after the switch. The unreported results indicate that a comparable extent of disclosure activity after switching to SETSmm affects spread in a more pronounced manner relative to the period before.

²¹The estimate for the combined effect of disclosure and trading (from the fixed effects specification) that is computed as the sum of the coefficients on DIS and DIS*SET is −0.105 (i.e. 0.007–0.112) with a p-value of 0.05 (not reported in Table 4).

²²In unreported analysis which re-estimates model (2) by replacing levels in SPREAD and DIS with their respective changes, we confirm the reinforcing or multiplicative association between disclosure and hybrid type of trading.

²³Our results continue to hold if we include the life cycle proxies and their interactions with DIS and DIS and SET, respectively.

²⁴This result is consistent with the concept that for small and less frequently traded stocks, changes in the disclosure strategy might actually increase volatility and illiquidity (Leuz and Verrecchia, 2000).

²⁵The estimate for the combined effect of disclosure and trading computed as the sum of the coefficients on DIS and DIS*SET has a p-value of 0.001 (not reported in Table 4).

²⁶The requirement for comparable distributions of disclosure level in quarters before and after the switch is appropriate for the regression analysis that tests the strength of the impact of disclosure on spread pre- versus post-switch. This requirement is now relaxed and the full range of sample observations is used in order to obtain a total number of frequencies of disclosure improvements.

²⁷The results for the switching group (adopters) are not reported as they are not materially different from those obtained estimating model (2) and reported in Table 4.

²⁸The test of the significant difference between the coefficients on DIS across the two classes of firms using the seemingly unrelated regression model is based on the Chow test (not reported) with a p-value of 0.273.

²⁹Our results remain robust if we include turnover growth as a life cycle proxy and its interactions with DIS and POST, respectively. Hence, taken together with the results concerning adopters (see Endnote 19) these findings alleviate concerns that potential life cycle differences between adopting and non-adopting groups might affect our original results.

³⁰We obtain almost identical estimates of the slope and intercept coefficients for the adopters sample if we employ model (2a) instead of (2). In other words, if we replace ‘adopters before (after) switch’ with ‘adopters before (after) 2005q4’.

³¹The direction of the bias depends on the response variable (market outcome) in question. For example, consider a scenario where a thinly traded risky start-up AIM firm chooses not to publicly disclose information on a new project in the pipeline due to high proprietary costs. Thus, ignoring factors (such as the riskiness of operations) that determine a firm's disclosure activities and also affect its bid–ask spread would yield a downwards bias in the OLS coefficients that exaggerate the effect of disclosure on the bid–ask spread.

³²An extended discussion of the empirical tests that assess the impact of lagged (in addition to contemporaneous) performance measures on the bid–ask spread are available in the online version of the paper at www.ssrn.com.

³³Note that although the focus of the paper is on the interaction between DIS and SET, the purpose of the analysis in Section 5.4 is to check for the endogeneity of the disclosure variable and to validate the method's approach used in models (1) and (2).

³⁴The R-squares of the first-stage regression indicate the strength of the correlation between the endogenous variable in question and the set of instruments. Higher values indicate stronger instruments, and instrumental variable estimators exhibit less bias when the instruments are strongly correlated with the endogenous variable. If the correlation is weak, then the 2SLS approach can produce biased estimates if the instrumental variables are even slightly endogenous (Larcker and Rusticus, 2010).

³⁵In this test, the null hypothesis is that the instruments are uncorrelated with the disturbance term ϵ from the Equation (1). If this hypothesis is rejected, one or more instruments do not appear to be uncorrelated with the error ϵ and are deemed endogenous.