We have shown that aggregate stock market volatility has been quite sta- ble over time, whereas f irm-level volatility has trended upwards. An impli- cation of this f inding is that there should be a declining trend in the correlations among individual stock returns. Declining correlations allow the volatility of the market portfolio to remain the same even if there is an increase in each individual stock’s volatility.We document the evolution of correlations among individual stocks by cal- culating all pairwise correlations among stocks traded on the NYSE, AMEX, and Nasdaq.5 We use both daily and monthly data. Correlations using daily data are calculated each month, using the previous 12 months of daily ob- servations ~or as many months as are available at the beginning of the data set!. The number of stocks in the sample at each month ranges f rom about1,500 to about 8,000, so the number of pairwise correlations ranges f rom just over 1 million to 32 million. We calculate an equally weighted average of these correlations. Correlations using monthly data are again calculated each month, but they use the previous 60 months of monthly returns. Somewhat fewer stocks have a complete f ive-year history than have a one-year history, so the number of stocks in the sample ranges f rom 1,000 to 4,500, and the number of pairwise correlations f rom half a million to just over 10 million. Again we calculate an equally weighted average for the month. The results are reported in the top panel of Figure 5.The f igure shows a clear tendency for correlations among individual stock returns to decline over time. Correlations based on f ive years of monthly data decline f rom 0.28 in the early 1960s to 0.08 in 1997, and correlations based on one year of daily data decline f rom 0.12 in the early 1960s to be- tween 0.02 and 0.04 in the 1990s. The former correlations are larger than the latter, both because daily stock returns contain negatively autocorre- lated idiosyncratic components, and because correlations are lower in more recent data, which receive greater weight in the daily calculation.
The bottom panel of Figure 5 plots the average R 2 statistic for the market
model, using the same stocks as the top panel. For each stock, the market model is estimated using f ive years of monthly or one year of daily data, and using the NYSE0AMEX0Nasdaq composite index as the market index. The resulting R 2 statistic is averaged across stocks. The two panels of the f igure are almost indistinguishable f rom one another. This is not surprising; if all
đang được dịch, vui lòng đợi..