I have had a number of discussions with investment managers who seem to be skeptical of my assertion that stocks and bonds have a very negative correlation on a daily basis. This fact is necessary and used by my gauges to determine prospective bond performance. If the S&P is out of gas, that is bullish for bonds only if there is a negative (short term) correlation. This component of the bond gauge will be a big problem if the two series are positively correlated.
Let’s start this by looking at cumulative gains in bonds when the S&P falls by at least .2 ATR’s. I need to set a threshold to eliminate noise from flat days. I’ll add the bond gain when the S&P falls and subtract the losses when the S &P rises.
As expected bonds went up when stocks fell by a greater amount than they fell when stocks rose. That is a by-product of a bull market data set. Nevertheless, bonds clearly fell by a considerable amount and very reliably so, when stocks rose.
We need to look at the 2004-2010 period as well to check for smoothness and performance when bonds were going nowhere.
During this very dull period ee see that the relationship was weak but this may be explainable (bear with me).
As we know, bonds and stocks have been generally rising for thirty-five years so many long term correlations will look positive. This is the source of the confusion. If there truly were a positive correlation, then any rally in equities would not hurt bonds. We can look at weekly data to see if the reliable (daily) negative correlation breaks down:
This is a little surprising – in the weekly time frame the negative correlation is even more pronounced! We used a higher filter here (.5 ATR’s) because the S&P has 5 days to move around. The other surprise is that bonds are just as likely to fall when the S&P rises as they are to rise when the S&P falls. Given the upward skew of the data set, that is a hard thing to do. The only hiccup in this virtually perfect relationship occurred in June 2013. This is important because the lookback periods in my gauges are long enough to need this relationship to remain intact for weekly (holding) periods.
How did this look in the 1997-2005 period?
In particular, during 1999 we had a severe downturn in bonds as the Fed tried to stall the Nasdaq bubble. In June of 1999 bonds fell regardless of what equities did. That is the kind of situation that would adversely affect signals I generate where the S&P condition is important. The other factor that I care about Is inflation and oil prices are a major factor. From April ’99 to October ’99 oil rose almost 50% so my bond gauge would have been very unhappy even though equity prices were not providing us with useful information (negative beta).
We can see that the relationship still held for the most part. Inflation explained the two marked declines.
Both Factors Must Be Considered.
As long as inflation explains divergences between bonds and stocks that are inconsistent with the traditional negative correlation then our approach will work. We must be worried if (say) bonds declined while inflation and stocks both fell. This may happen but it is historically extremely rare. During the low volatility period of ’05-’07 we saw a slight decline in the sum of bond price gains, attributable to S&P movement (see figure 2). This is partly to the fact that such low volatility will naturally reduce elasticity but there is a second reason: When S&P volatility is very low inflation volatility may become the dominant factor, just as it was in June ’99 and during that ’05-’07 period.
You can see here that oil volatility was indeed consistently higher than S&P vol. so our inflation component dominated and should help us account for bond price movements under these conditions.
We will continue to assume that if the stock market declines precipitously bond prices will rise or fall very little, unless oil and other inflation barometers are rising simultaneously. The traditional flight to safety has been true essentially forever (just like it says in all the textbooks we read) and we see no reason for it to stop.