Thin markets are characterized as markets with a low number of buyers and sellers. Since few transactions take place in a thin market, prices are often more volatile and assets are less liquid. The small number of market participants results in low transaction volume and relative liquidity. The same logic can be applied to a single stock. There are several stocks available on the market with different properties. The characteristic of low trading frequency is commonly observed with less known companies listed on the market. Estimating betas for companies with less-liquid stocks proves to be a tad more challenging compared to estimation of large high-liquid stock.
Problem of liquidity – Thin trading
Thin traded stocks usually describe shares with not readily available information about their price and shares that are not frequently traded. It is clear that new information regarding such companies may lead to non-synchronized trading behavior, because fewer people trade and they may not be quite so well informed, as people participating in more liquid markets. Lag between arrival of information and the trade results in reduced relationship between returns of market index and stock price. Consequently beta, as the measure of co-movement, is smaller, which in turn leads to a biased estimate of a company value. The next segment examines how a substantial amount of trading periods with zero returns effect the estimated betas.
We start the procedure with simulating 248 observations of returns of arbitrary stock X and market index Y. Returns are simulated based on some assumptions regarding the mean, standard deviation and correlation between the two series. Based on the assumptions about parameters of the two series of returns, anticipated value of estimated beta is around 0.84. In the first case the analyzed stock is not affected by thin trading’s zero returns. First example is presented below.
The regression estimation produces a beta estimation of 0.85 which is very close to the predicted estimate.
Second example considers a case where thin trading is present. Once more returns were simulated with the same assumptions regarding parameters that describe the two series. But this time 50 observations were chosen at random and we assign them zero return to illustrate low frequency trading stock. Essentially we consider a stock that is not traded on a continuous basis, and for such assets beta estimations can be affected. In particular, non-trading on an asset during a return period can reduce the measured correlation with the market index, and consequently beta estimate. The plot below displays scatter plot and fitted regression line of the thin-traded stock.
As predicted, estimated beta (0.69) is considerably lower from the previous analysis. To show that the two estimated betas are not a consequence of some anomaly we extended the simulation to estimates of beta. 1000 betas were estimated based on 1000 sets of simulated returns with the same underlying assumptions. Histogram below represents the results of the process.
It is clear that the stock with characteristics of low frequency trading displays significantly lower estimates of beta on average.
There are techniques to obtain unbiased beta estimations in infrequently traded environment. The simplest way is to consider longer return interval. For example using monthly returns instead of weekly returns can reduce the number of zero return periods. Another possibility is to exclude those periods from the analysis in which we do not observe trades.