Page 3 of 7
Why does measured correlation differ from its long-term average? The fact that observed correlation varies, even over relatively long periods of time, does not necessarily mean that "correlations are changing," although this may be the case. It simply reflects randomness in the return variables themselves, which generally produces ex-post outcomes that differ from the "true" underlying statistic or longer-term average, particularly over shorter periods.
Previous research suggests that not only does randomness affect measures of realized correlation through time, but also that the underlying correlations between asset returns change over time and in particular circumstances, and have important relationships to events such as volatility shocks. Ilmanen  found that factors increasing the correlation between U.S. stocks and bonds include high inflation and significant changes in GDP growth. Ilmanen also found that stock-bond correlations tend to be lowest when equities are weak and volatile, such as during flights to quality. Other research has provided similar evidence. Gulko  found that stock-bond correlations are positively related during normal market conditions, but decrease during stock market plunges. Connolly et al.  showed that stock-bond correlation is lower when the implied volatility from equity index options is higher.
Although market volatility has emerged as a key driver that tends to decrease correlations between stocks and bonds, volatility is also a major driver that tends to increase correlations when looking at subcomponents of the same asset class. For example, numerous studies have found that correlations between U.S. and international stocks increase substantially during volatile market episodes.6 Longin and Solnik  found that correlation is not related to market volatility per se, but to the market trend, with correlation increasing during bear markets but not in bull markets.7
Implications For Portfolio Construction