Constructing Market Cycle and Macro Regime Indicators

My cycle and regime indicators are based on the idea of similarity. For instance, if we are trying to understand if we’re going to enjoy a particular new cocktail, we’ll compare it’s ingredients to other cocktails we know that we like. So chances are if you like a Manhattan which is whisky and sweet vermouth, you’ll like a Boulevardier which is whisky, sweet vermouth and Campari. Maybe not, but the similarity between the two drinks is so high that the odds are you’ll like it. I know I do.

Constructing market state indicators is much the same, though there is more uncertainty in them because, unlike the cocktails, we don’t know all of the ingredients for something like a recession. To use another analogy, it would be similar to the methodology ancient people would have used to predict the seasons. By “ancient” here I’m talking about before civilization when we roamed in tribes hunting and gathering. Back then we didn’t have calendars and definitely didn’t know that seasons were caused by the tilt of the Earth towards the sun. So instead we would look for signs that winter was coming. Things like leaves turning color, the days feeling colder, the sun being lower on the horizon, and birds flying south. Any one of these signs without the others wouldn’t mean anything. But as more and more of them built up, our ancestors would think winter was coming, so its time to prepare. The signs they used were actually leading indicators, and as more of them turned on the preponderance of the evidence would point to a seasonal change. While ancient people didn’t know the causality behind seasonal change they could observe the symptoms of such change and so tell one state from another.

Both the cocktail analogy and predicting the change of seasons are how we intuitively use fuzzy sets to deal with uncertainty. The cocktail example is a static object while predicting the seasons is dynamic, but the both use the same basic approach. How similar is the unknown to the known? Note that this is not probability because probability depends upon frequency. That is repeating the event and figuring out how often your prediction is right. With the cocktail we can say that the Boulevardier is 75% similar to a Manhattan. So you might think that there’s a 75% chance we’ll like the Boulevardier. But once we taste it we’ll know 100% whether we like the Boulevardier or not (assuming you’re an all or nothing type of person when it comes to drinks), but its ingredients will remain 75% similar to a Manhattan. Similarity does not change with sampling unlike probabilities because it’s a state measure, not a measure of chance.

Signal Methodology

To use fuzzy set principals we need to compare present conditions to past occurrences of the event. For instance, if we wanted a model for anticipating changes in central bank policy we’d look at what types of conditions existed in previous times when the central bank has raised or lowered rates. There could be something like a rise in inflation expectations and high industrial production. These factors would be turned into binary signals based upon thresholds. In the past, when Industrial production was above 75% in the US the Fed would raise rates, for instance. The signals themselves should work as stand alone leading indicators, but sometimes one works, sometimes another and sometimes they both work. By combining them together you should get a better forecast. Unlike regression based forecasting methodologies, this approach can use highly correlated data. So the more signals you have to combine into an indicator the better, though you have to be sure that combined together they work better than they do separately. Sometimes factors combined together are worse than they work independently. There is an entire literature on binary signals.

Analyzing each binary signal’s threshold is based upon a technique called ROC analysis. ROC stands for “Receiver Operating Characteristic” and was originally developed in World War 2 for radar analysis. Basically, its a way to determine whether you’re making a Type 1 or 2 error. By using many samples you can find at what level the indicator reaches maximum accuracy. As a freebee I can say that the threshold for the Markit PMI to predict an economic slow down is not 50 but 52. The Markit Purchasing Managers Index (PMI) is a survey of businesses on current business conditions. The answers to the survey are complied so that when the PMI is above 50 it means that more managers think that conditions are improving, and when its below 50 conditions are deteriorating. For this reason the interpretation of the PMI typically uses 50 as a threshold. But the ROC analysis showed that slowdowns in GDP were more accurately predicted when the PMI was below 52. This likely reflects the human tendency to be overly optimistic most of the time. So in my indicators when the Global PMI drops below 52 it means we’re heading into a manufacturing recession. Likewise the PMI has to rise to above 52 to signal recovery and renewed growth. So using ROC analysis we can determine the proper thresholds to construct binary signals.

Using Principals of Concordance we can determine how similar the signals are to each other. That is, how often do they agree that the event is going to happen.

Creating Indicators

Combining the signals does not depend upon fixed weights. It is based upon the similarity, using fuzzy set principals, of the current set of signals that are on or off to past states. Similarity is based upon a membership function that ranges from 0 to 1. I must emphasize that this is not a probability though it looks like one because it uses the same scale. Remember the 75% similarity between the Manhattan and Boulevardier did not change once you sampled it, though your probability of liking it changed from 75% to 100% or 0%. The resulting indicator gives us a read of the current state and we can use that information to see how markets will react. For instance, there has been a realization that high levels of inflation in 2022 have caused stocks and bonds to become more highly correlated with one another so that bonds became a less a reliable hedge for stocks. In the newsletter section of this site you will find papers I published several years back that had already found this to be a likely consequence of high inflation.

Conditional States

Since all the systems are interrelated, various combinations also tell us something. For instance the stock market can be in a state of high uncertainty but still have good returns if leverage is low. However, if the financial system is unstable due to high leverage and the stock market is unstable as well, we are more likely to have a financial crisis such as 2008. That’s why I publish the four indicators together though the Regime indicators change very slowly compared to the cycle indicators. But the combination is quite important.

Conclusion

This article was to introduce the ideas behind the methodology. In the other entries I will discuss the signals that go into each of the four indicators and what they mean.