Idiosyncratic risk (IR) plays a central role in financial markets. There is an extensive literature (reviewed in the next section) trying to understand the determinants of time-series variation in average idiosyncratic risk. This literature is especially focused on understanding the high average IR in the late 1990s and the overall increase in average IR from the 1960s to the early 2000s. To date, there is no consensus in the literature. In this paper, we show that IR is historically low over the recent years and explore this empirical fact to shed light on the determinants of IR variation. Our key result is that the low IR of the recent years can be explained by variation in the characteristics of listed firms. We argue that such variation is linked to the dramatic changes in the number and composition of public firms that have taken place since the late 1990s.
The literature has provided several explanations for variation in IR, and they can be broadly classified as based on (i) macroeconomic variables (including market risk) and (ii) firm characteristics. We follow this classification and explore both as possible explanations for the low IR of the recent years. In the first part of the paper, we show that macroeconomic models of IR predict a lower than historical average IR for 2013-2017, but they cannot explain why IR is so low. While there is a strong relation between market risk and IR, market risk in recent years is not significantly lower than most years in our sample period. Using the historical relation between IR and macroeconomic variables from 1963 to 2012 predicts equally-weighted IR to be 31% from 2013-2017 when actual IR is 26%, and predicts value-weighted IR to be 20% when actual value-weighted IR is 17%. Adding policy uncertainty, which is known to affect firm volatility (Baker, Bloom, and Davis, 2016), to the macroeconomic variables does not help reduce prediction errors.
In the second part of the paper, we focus on firm characteristics. Small and young firms are known to have much higher IR. High R&D expenditures, high leverage, low profitability, and less liquidity are also associated with high IR. The recent literature has shown that the number of listed firms falls dramatically since the late 1990s, so that the number of listed firms in the United States is now less than in the 1970s (Doidge, Karolyi, and Stulz, 2016). Moreover, the listed firms in recent years are quite different from the listed firms in the 1990s (Doidge, Karolyi, and Stulz, 2016; Kahle and Stulz, 2017). Importantly, firms are larger, older, and more profitable; their common stock is more liquid. Such substantial and systematic changes in the characteristics of listed firms imply economically large decreases in IR relative to the late 1990s.
Since the characteristics of listed firms change dramatically over time, we use a novel approach to explain IR based on its link to firm characteristics. Specifically, we estimate models using the sample until 2012 and use the model slopes to predict idiosyncratic risk at the firm level from 2013 to 2017. We then average predicted firm-level IR to obtain average estimates of IR. This approach reflects the distribution of firm characteristics each month and hence takes fully into account changes in firm characteristics as long as the determinants of IR at the firm level are reasonably stable, which we show appears to be the case.
Using firm characteristics to explain firm-level IR and aggregating the results across firms explains the low recent IR much better than using only macroeconomic variables or using time-series regressions where the dependent variable is average IR and explanatory variables are macroeconomic variables and average firm characteristics. When we add lagged values of book/market, firm age, and firm size to a panel regres-sion that has only lagged macroeconomic variables, predicted equally-weighted IR is 27% compared to actual IR of 26%. With value-weighting, predicted IR and actual IR are both 17%. In other words, firm characteristics can fully explain why value-weighted IR is so low in the recent past.
We examine how our results are consistent with evidence that market risk is not historically low in recent years. If IR falls but market risk does not, it must be that stock returns become more correlated with the market. Large firms have higher market-model R-squareds. Hence, changes in firm characteristics should also lead to an increase in R-squareds. In line with this prediction, we find that average R-squared from 2013 to 2017 is higher than in any year from 1963 to 2006.
The next section provides a summary of the related literature. Section 2 describes our data and con-struction of risk measures and other variables. Section 3 studies idiosyncratic risk over time. Section 4 investigates whether macroeconomic variables can explain the low recent IR. In Section 5, we show that firm characteristics have more explanatory power for IR than macroeconomic variables and that they can explain the low recent IR. Section 6 shows that market risk is not unusually low in recent years and that market model R-squareds have increased as expected given the changes in firm characteristics. Finally, Section 7 concludes.