Classifying Economic Regimes

A data-driven approach to clustering similar macroeconomic environments
 

By: Chris Satterthwaite, Lionel Smoler Schatz, and Oleg Laskov

Last week, we discussed how to integrate macroeconomic signals to measure the relative similarity of macroeconomic climates across time.

Our next challenge is to attempt to group different periods into economic regimes to help us understand current conditions within a historical context.

To do so, we applied the Euclidean distance concept outlined last week to daily macroeconomic data from 1962 to 2024. We used an unsupervised clustering algorithm called a Gaussian mixture model (GMM) to fit distributions to our historical macro signals.
Below is an illustrative visualization of how this method works using simulated data.

Figure 1: Illustrative GMM Regime Classification

Source: Verdad analysis

In the illustrative example above, the model finds two clusters of economic regimes: one defined by low GDP growth, low inflation, and an upward-sloping yield curve (blue), and another defined by high GDP growth, high inflation, and an inverted yield curve (orange).

Applying GMM to all our macro signals from 1962 to 2024 reveals four distinct economic regimes:

• Regime 1: Defined by benign economic conditions—moderate rates, upward-sloping yield curve, subdued volatility, etc. We call this regime Growth.

• Regime 2: Defined by high inflation, elevated rates, and high bond volatility. We call this regime Inflation.

• Regime 3: Defined by risk-taking behavior (compressed high-yield spreads), depressed volatility, a flat or inverted yield curve, and increasing stock-bond correlation. We call this regime Precarious, as it tends to precede crises.

• Regime 4: Defined by high risk aversion and significantly elevated volatility. We call this regime Crisis.

The figure below shows the average standardized signal in each regime, allowing us to visualize the defining characteristics of each regime, as mentioned above.

Figure 2: Regime Characteristics

Source: Verdad analysis

This exercise and the resulting four regimes are a more quantitatively rigorous extension of the economic quadrant framework we laid out in our research piece on Countercyclical Investing, which builds on the research of Bridgewater and Hedgeye. While our previous quadrant framework was more prescriptive than the GMM analysis, the findings were surprisingly similar.

Figure 3: Countercyclical Investing Economic Quadrant vs. Regime Corollary

Source: Verdad analysis

Equipped with these economic regimes, we can classify each time period into one of the four and examine how the economy has cycled through these regimes over time.

Figure 4: Economic Regimes Over Time (1970-2024)

Note: The chart above shows the broad regimes at each time period. There are many transitions within regimes that are too granular to show here.
Source: Verdad analysis

The graph above seems to corroborate our classifications. The crisis regime correctly captures the inflationary 1970s, the tech bubble in the early 2000s, the financial crisis of ’08, and, most recently, COVID and the inflation spike that followed.

As far as our current macro environment is concerned, we fit most closely with the precarious regime, similar to early 2007. This is consistent with our first similarity analysis using Euclidean distance, as we discussed last week.

While economic regimes can help us classify similar environments based on different macro signals, they can also offer insights when looking at returns. By looking at the behavior of different asset classes and factors within the four regimes, we can think about what each regime implies about asset class returns. Below we show the average annualized return for various assets and factors in each regime.

Figure 5: Asset and Factor Returns by Regime (1970-2024)

Source: Verdad analysis

Briefly, we’ll walk through the relative best and worst performers in each regime:

• Regime 1 (Growth): equities, corporate bonds, and oil all do great. Smaller and cheaper companies do better here as well. Equities also did best in Quadrant 1 in our Countercyclical Investing paper.

• Regime 2 (Inflation): Oil excels here, along with the S&P 500. US Treasurys do poorly, and most equity factors don’t work great. We similarly found that equities and commodities did best in Quadrant 2 in our Countercyclical Investing paper.

• Regime 3 (Precarious): This is a bit of a mix, but most stuff does okay. Corporate bonds and gold do best. Gold was the second-best performer after energy in Quadrant 3 of our countercyclical framework.

• Regime 4 (Crisis): US Treasurys and gold do best. Oil and equities do terribly. Within equities, large companies with high profitability and low investment do better. US Treasurys were the best performer in Quadrant 4 of our countercyclical model.

Knowing what economic regime we might be in doesn’t give us a crystal ball, but it does help us aggregate disparate economic signals into a framework, which in turn allows us to make relatively stable generalizations about expected asset class and factor returns.

This can be a helpful tool for dynamic portfolio allocation, ensuring investors aren’t simply owning what worked over the prior decade but also what will work in the next.

Acknowledgment: Our intern Oleg Laskov has spent part of his summer on this research. He is a rising junior at Yale University majoring in applied mathematics. He is interested in quantitative finance, machine learning, and trading commodities.