Intrigued by the ideas presented in the Goldman Sachs research publication ‘Bear Necessities’, which presents a model using macroeconomic indicators to measure the likelihood of a bear market, we decided to take the research a step further.
In the Bear Necessities paper, Z-scores1 for the following five variables are calculated and combined into a recession indicator: unemployment, ISM, inflation, yield curve and valuation as measured by Shiller’s CAPE. Unfortunately when using the Bear Necessities approach, the indicators often reach the recession levels too early, which causes potential investors to miss out on too much of the run-up towards the recession to be of practical use.
Our goal was to find macro economic indicators that show a longterm cycle similar to that of the market. ISM and inflation turned out too noisy to be of any value. Shiller’s CAPE did not look very promising either as sell signals generated from it were generally too late to be of any use. Instead, we found an additional variable with long term cyclicality – the delinquency rate of commercial and industrial loans. Care has to be taken about the release dates of the data. For our analysis we only use point-in-time data.
The charts below show an inverse relationship between the cycles of the market and the cycles of our variables. When the market peaks, our variables are close to their lows. This means that if our variables are increasing the market must be going down and vice versa, if our indicators are decreasing, the market is going up. A natural way to measure this is by looking at the sign of the change of the variables. There might be short term fluctuations in the variables, which is why we decided to measure the change in our variables at the six month horizon. For delinquencies, which are released quarterly, this equates to the change between 3 observations and for unemployment, a monthly series, it equates to difference across 7 observations. The areas highlighted in blue are periods when the six-month change of the respective variable is positive and the green areas indicate a negative six-month change.
In the table below we show the average return for the times when the six-month change was positive and negative for the period between June 1988 and December 2019.
Delinquencies and unemployment indicators overlapped for the most part with the term spread indicator which is why we decided to exclude the term spread from our final model. In the final model, we are invested in equities whenever either of the indicators is negative and hold the risk-free asset whenever both indicators are positive.
What we like about the unemployment and the delinquencies indicators is the economic interpretation. The indicators tell us whether their underlying series are declining or increasing. Thinking about unemployment, one would think that during an economic boom new jobs are created and companies will lay people off during recessions due to lower demand for goods and services. Similarly, it is easy to imagine that during times of economic prosperity companies are able to generate enough revenue to make debt payments and during times of economic hardship, revenues are low and companies might default on their loans.
Critics might say that our results are not significant since our signal does not change direction often. We counter them with the simplicity of our model and the economic intuition behind it.
A Z-Score is calculated by taking a data point, subtracting the mean and dividing by the standard deviation. The Z-Score provides an Apples-to-Apples comparison of how far individual data points have moved away from their central tendency.↩