The Liquidity of the Two Share Classes of Dual-Class Firms: Implications for Asset Pricing and Microstructure.

Premal P. Vora[*]

Associate Professor of Finance

PennStateHarrisburg

School of Business Administration

777 W. Harrisburg Pike

Middletown, PA19335.

(717) 948-6148

John R. Ezzell

Professor of Finance

PennState

SmealCollege of Business

University Park, PA16802.

(814) 863-7969

JEL Classification Codes: G12, G30.

The Liquidity of the Two Share Classes of Dual-Class Firms: Implications for Asset Pricing and Microstructure.

Abstract

A solid body of evidence concludes that the liquidity of a financial asset has an impact on its price. Therefore, some of the disparity in price between the two share classes of a dual-class firm may reflect differences in their liquidity. We explore this topic by providing evidence on the magnitude of and the differences in the spreads, depths, and other characteristics of the two share classes of a dual-class firm. We find substantial differences in the liquidity of the two share classes. We also test models of specialist behavior that make predictions about how the bid-ask spread is related to share characteristics. We find support for most—but not all—models of specialist behavior. We also uncover some subtleties in how the specialist determines the bid-ask spread. Finally, we find that the extant models of spread determination are not able to explain all the difference in the spreads of the two classes of shares.

The Liquidity of the Two Share Classes of Dual-Class Firms: Implications for Asset Pricing and Microstructure

I. Introduction

In a frictionless capital market, the two classes of shares of a dual-class firm should trade at the same price. Both classes of shares represent claims against the same underlying stream of cash flows. In the absence of frictions, such as transactions costs, and informational asymmetries, arbitrageurs’ attempts to profit from any differences in price of the two share classes should restore the price parity of the two classes. However, in the presence of market frictions it is possible for disparities in price to persist for long periods of time. Such disparities reflect the presence of private benefits of control and how pivotal the marginal shares of the two classes are in order to get effective control (Zingales (1994), Zingales (1995)). The role of trading-related transaction costs in explaining disparities in price of the two share classes has largely been ignored in the literature. We take a first step in exploring this topic by providing evidence on the magnitude of and the differences in the bid-ask spreads of the two share classes of dual-class firms.

There is a growing body of literature that concludes that transaction costs play an important role in the pricing of financial assets. In one of the early contributions to this literature, Amihud and Mendelson (1986) demonstrate that in a capital market where the bid-ask spread varies across securities, in equilibrium the conditional return on each security is positively related to the size of its bid-ask spread. Further, their evidence based on NYSE-listed stocks strongly supports their model. Subsequently, Brennan and Subrahmanyam (1996),Eleswarapu (1997), Chalmers and Kadlec (1998), Datar, Naik and Radcliffe (1998), and Easley, Hvidkjaer, and O’Hara (2002) all provide evidence that less liquid stocks are priced to offer a higher rate of return. Further, the limits imposed by transaction costson arbitrageurs’ ability to profit from price disparities can also fail to bring about parity in prices, as shown by Garman and Ohlson (1981), Jones and Lamont (2002), and others.[1] Both of these lines of research raise the possibility that differences in returns (and in prices) of the two classes of shares of a dual-class firm could partly stem from the existence of the bid-ask spread andfromsystematic differences in the spreads of the two share classes. Therefore, we seek to provide evidence on the magnitude of the bid-ask spread for the two share classes of dual-class firms and on the systematic differences in the spread between the two classes.

The body of literature on liquidity and asset pricing takes the existence of the spread and its magnitude as given. However, it is also important to understand how the specialist determines the spread on any given security.This issue remains central to the study of market microstructure due to the implications it has for investor welfare, for market structure and design, for asset pricing, for public policy, and for managerial policies that may have an impact on the magnitude of the spread.[2]

Traders are willing to pay the bid-ask spread to a specialist because it is the price they pay to receive immediacy. If the market for providing immediacy is frictionless, the bid-ask spread on two securities that are perfect substitutes for one another should be equal; any difference between the spreads would be arbitraged away by other specialists seeking profit opportunities. The variance in spreads that has been found along a cross-section of securities in, for example,Stoll (1978b), Godek (1996), or Easley, Kiefer, O’Hara and Paperman (1996), therefore, must be related to the variance in the characteristics of securities that determine spreads. Demsetz (1968), Bagehot (1971),Tinic (1972), Stoll (1978a), Easley and O’Hara (1987) and others have created models of specialist behavior that have testable predictions for the relation between the spread and certain characteristics of shares—such as trading activity, price, volatility, and the presence of informed trading—that are posited to affect specialist behavior. Once all these characteristics have been controlled for, however, the spreads between any two securities should be the same.

While the concept of perfect substitutability is useful for making predictions about how markets will function, in practice it is virtually impossible to find non-trivial examples of perfect substitutes.[3] At the same time, it is possible to find many non-trivial examples of assets that are close but not perfect substitutes. One such example, of course, is that of the two share classes of dual-class firms. In addition to providing evidence on the magnitude of and differences in the spreads of the shares of dual-class firms, we use the special characteristic of close substitutability to test the extant models of spread determination. If the magnitude of the spread charged by the specialist is determined solely by trading activity, price, volatility, and presence of informed trading, there should be no difference in the spreads of the two share classes of dual-class firms once these factors have been controlled for. It is particularly easy to control for some of these characteristics for the share classes of dual-class firms because there is either no difference or the difference is small. Thus, this facet of dual-class firms provides us with a powerful opportunity to test the extant models of spread determination.

Our sample consists of 62 NYSE-listed dual-class firms in the period August 1999 to October 1999. Both classes of shares for the firms that appear in our sample have the same cash-flow rights, but differ in their voting power. Thus, every firm that appears in the sample has one share class that carries more voting rights per share than the other. We refer to the class with more voting rights as “high-vote stock” (HVS) and to the class with fewer voting rights as “low-vote stock” (LVS).

We find that HVS sells at a premium of 3.6% to LVS over the sample period, but the two classes are virtually identical in terms of return volatility. We also find that HVS trades less frequently than LVS and the lower trading frequency translates into lower volume for HVS. Further, we find that spreads—estimated from NYSE’s TAQ data—on HVS are significantly higher than they are on LVS. We also find that the spreads decrease in trade size and in firm size. We then decompose the effective spread into its two components—the realized spread and the adverse-selection component. We find that the realized spread for HVS is lower than it is for LVS. The adverse-selection component of the spread, however, is higher for HVS than for LVS. We also study the differences in quoted depths between the classes and find that depths are generally higher for HVS than for LVS. Further, we find that the presence of informed traders is higher in HVS than in LVS. Our results clearly raise the possibility that some disparity in the prices of the two classes of shares stems from the existence of and the differences in the bid-ask spread and depth of the two classes, either because of the impact of liquidity on asset prices or because liquidity-imposed barriers to arbitrage fail to bring about parity in prices.

When we cross-sectionally relate the spread to share characteristics we find that our results are consistent with many—but not all—features of models of specialist behavior related to the spread. We find evidence that order-processing costs are fixed with respect to size of order and that inventory-holding costs are related to the return volatility of shares. In our regressions relating spread to proxies for the presence of informed trading we find that the two are positively related for LVS but unrelated for HVS. In light of the evidence that HVS has a higher presence of informed traders and that both the effective half-spread and the adverse-selection component of the effective half-spread are higher for HVS, it appears that the specialist adjusts the spread for the probability of informed trading only for LVS but keeps the adverse-selection component of the spread high and constant in HVS ignoring any variations in the presence of informed trading in HVS. To our knowledge such subtle but important distinctions in how the specialist determines the spread have not been found in any previously published research.

We also find that spread is unrelated to the price of the stock. Thus, our evidence finds no role for the price of the stock in the specialist’s decision process. Finally, despite the close fit between the dependent and independent variables in our regressions, we consistently find that a statistically significant difference in the spreads of HVS and LVS remains unexplained by the selected independent variables. This suggests the possibility that the extant models of specialist behavior are not able to capture all the economic factors that might be relevant for setting the bid-ask spread.

II. Data and Methodology

We identify, from proxy statements filed with the Securities and Exchange Commission, all dual-class NYSE-listed firms with equal cash flow rights that were publicly traded throughout August to October 1999. Our sample consists of 62 firms. We were able to retrieve trades and quotes for both classes of shares for each firm from NYSE’s TAQ database. Additionally, we retrieved daily returns from CRSP for the period August 1999 to October 1999 and Income Statement and Balance Sheet items from Compustat for fiscal years 1998 and 1999.

In Panel A of Table 1 we present summary statistics on the differences between the two classes of shares of our sample firms. In Panel B we compare our dual-class firms to NYSE-listed firms that are not in the dual-class sample. In Panel A, we begin with the ratio of the price of HVS to the price of LVS. For each firm we calculate the average price of HVS over all trades during the sample period and divide it by the average price of LVS over all trades. This number is then averaged across all firms. From the mean of this ratio, it appears that HVS carries a statistically significant (at the 1% level) price premium of 3.6 percent over LVS. However, the considerable variation in the ratio across the sample suggests that not every HVS sells at a premium to its LVS. This variation is apparent from the minimum (0.842), the maximum (1.391), and the median, the 25th and the 75th percentile values for this ratio that are reproduced in the table. The magnitude of the mean premium and the fact that there is substantial variation across the sample are both consistent with prior studies (Lease, McConnell, and Mikkelson (1983), (1984); Zingales (1995)).

[Place Table 1 here]

Based on the results appearing in the next four rows, we find that HVS trades less frequently than LVS and this translates into lower monthly volume for HVS. On average there are 589 trades per month in HVS with an average monthly volume of 1.53 million shares; for LVS there are 1,206 trades per month with an average monthly volume of 3.68 million shares. Both differences are statistically significant at the 5% level based on the ordinary t-test.[4] When we compare the two classes of shares in terms of their return volatility as measured by the standard deviation of daily returns in the period August 1999 to October 1999, we find them to be virtually indistinguishable. Thus, while HVS is traded less frequently than LVS, the two appear to be identical in terms of daily return volatility. Next, we characterize the voting power, management ownership, and the market capitalization of the two classes of stock. Voting power and management ownership are collected from proxy statements while market capitalization is the market value of that class of equity at the end of October 1999. The voting power of HVS ranges from a minimum of two-thirds of a vote per share to a maximum of 20 votes per share with a mean of 2.84 votes per share. For LVS, the minimum is zero votes per share, the maximum is one vote per share with a mean of 0.264 votes per share. On average, LVS has a statistically significantly different one-tenth the voting power of HVS.Further, the ownership of HVS is concentrated with the management of the firm as, on average, management owns 46% of the outstanding shares. Management owns a statistically significantly different only 23% of LVS. Finally, while the market capitalization of HVS appears to be smaller than that of LVS, statistically they cannot be distinguished from one another.

In Panel B, we compare the size and leverage of dual-class firms to all NYSE-listed firms that do not appear in our sample. The typical dual-class firm, with Total Assets of $4,561 million, is one-third the size of other NYSE-listed firms in terms of Total Assets and, with Sales of $2,065 million, a little less than half the size of other NYSE-listed firms in terms of Sales. Both differences are statistically significant at the 1% level. These differences in size are qualitatively comparable to the difference found, for instance, by Partch (1987), in the market value of the stock of dual-class firms and non-dual-class firms. Finally, although dual-class firms are smaller, they are levered no differently than their larger counterparts with a debt ratio of 25.71 percent.

We now turn our attention to the microstructure differences between the two share classes of these firms that requires a thorough analysis of the transactions data. We apply several screens to the TAQ data in order to eliminate any errors that might be present. We retain only the national best bid and ask quotes. Any quote where the bid  ask is eliminated. All bid, ask and trade prices must be positive. All quotes and prices must be divisible by 16. Only quotes with positive depth are admitted. Any quote where the ask is more than ten percent over the bid is eliminated as an error. Any out of sequence trades or trades with special settlement conditions are eliminated. We also eliminate all trades that occurred before 9:30 am and after 4:00 pm. We match trades with the first quote at least five seconds prior as suggested by Lee and Ready (1991). If no quote is available prior to the first transaction of the day it is eliminated. We also cumulate and treat as one trade any consecutive trades during a day with the same price, bid price, and ask price. After applying the matching and cumulating algorithm to our data, we obtain a total of 320,792 trades across both classes of shares; of that, 103,663 trades are for HVS and 217,129 are for LVS. In the next section we present and discuss our results based on the analysis of this data.

III. Results

We begin by providing evidence on the magnitude of the quoted and effective half-spread and quoted depths. The quoted half-spread at time t is defined as

Quoted half-spreadt = (Askt – Bidt) / 2*Mt,

where Askt and Bidt are the ask and bid prices prevailing at time t, and Mt is the midpoint of the bid and ask prices at time t. In order to facilitate comparisons across stocks that are traded at different prices, we report spreads throughout the paper as a percentage of the quoted midpoint. If a transaction occurs at the ask price or at the bid price, the quoted half-spread is a measure of the one-way spread-related cost of the transaction. Our results on the quoted and effective half-spreads appear in Panel A of Table 2. On the left-hand side of Panel A, we display our results on the quoted half-spread while on the right-hand side we display our results on the effective half-spread.