Regimes, Revisited

There was a time when the word “regime” conjured images of French dynasties rather than financial markets. Then Covid happened. The inflationary shift that followed brought the concept of market regimes from the academic fringes to the heart of investment strategy. It became clear that the standard calendar-based approach to market analysis, anchored in months, quarters, and years, was an artificial constraint (albeit, as our trend and momentum strategies make clear, a useful one). Returns don’t always adhere to the Gregorian calendar, though. What matters are the underlying forces shaping asset prices at any given moment.

A regime-based framework acknowledges this, grouping together periods when financial conditions exhibit deep underlying similarities rather than adhering to arbitrary dates. Once you start thinking in terms of regimes, it changes everything. Regimes provide a framework for understanding why a strategy that worked last year may no longer be valid today. They force us to confront a simple but essential truth: markets evolve, and so must our models and our mindsets.

At Man Group, we’ve been studying regimes for years. In 2019 and 2020, before inflation was on anyone’s radar, Henry Neville , Otto Van Hemert and others did some deep research on the subject. That research helped us anticipate and navigate the 2021-2022 inflation shock. Now, our latest work takes the concept of identifying and responding to regime shifts even further.

In this paper, written by teams from our London and Boston offices, led by Amara Mulliner and the great Campbell Harvey , we seek to shape a new way of thinking about regime-based investment. Rather than defining regimes in advance - “high growth,” “low inflation,” “monetary tightening” - we let the data tell us which past environments most resemble today’s. The underlying intuition is similar to a k-Nearest Neighbours (k-NN) algorithm, a staple of machine learning. Just as k-NN finds the closest historical data points to classify an unknown observation, our model searches for periods in which financial conditions were most similar to the present. If markets behaved a certain way in those past regimes, they are more likely to behave similarly today.

To do this, we tested over a hundred economic indicators and identified seven key state variables as fundamental drivers of stock returns.

Each variable is transformed into a standardised z-score, and we systematically search history for periods where these variables align most closely with their current state. If the past returns following these historical analogues were positive, we take a long position; if they were negative, we move defensive. It’s simple, but it’s effective.

?Even using a relatively simple version of this approach, the results are striking. The model systematically identifies periods of regime similarity that contain meaningful predictive power.

The chart below shows in practice how this model would have worked for a date in the GFC – enabling us to look at similar periods in history for a guide as to how we might expect markets to respond. The lower the Global Score (which is just the combined z-scores for our indicators), the nearer to the current period a historical period is. We have outlined in blue the 15% most similar months. So we can see that the bursting of the dot com bubble and the recession of the early 1980s were both similar to the GFC, with the dot com bubble exhibiting a number of such similar episodes.

There’s information to be gleaned from the steepness of the line, too. The steeper the line, the less similar the current period is to historical periods. We can use this to understand when a period might represent a change or inflection point in the market regime. The chart below uses our model to identify change points, something it appears to do with some measure of success. Our researchers at the Oxford-Man Institute are doing some further work on this, looking at how machine learning can help refine and develop this identification of regime shifts.

The power of a regime-based framework is that it forces us to think about markets in terms of structural shifts rather than arbitrary time periods. And one of the defining characteristics of regime shifts is that they are rarely smooth.

Volatility is not just a side effect of regime change; it is a signal that the market is repricing risk in a fundamentally different way. Consider the inflation shock of 2022. For over a decade, markets operated under the assumption that central banks would swiftly intervene to suppress volatility. That assumption collapsed when persistent inflation forced an aggressive tightening cycle. The resulting regime shift was marked by extreme cross-asset volatility: stock-bond correlations flipped, long-standing risk premia reset and asset allocators were left navigating an entirely new investment landscape.

This is not an isolated example. Stability begets complacency; when market conditions change, volatility erupts as investors scramble to recalibrate expectations. Recognising these transitions in real time is critical, and a regime-based framework provides precisely that lens.

Once you start viewing markets through the lens of regimes, a broader sphere of research opens up. We can ask better questions. What are the natural regimes for currencies, and how do they shift? How should investors adjust their allocations when we transition from a monetary easing regime to a tightening one? What does it mean for factor investing if the relationships between styles and macroeconomic conditions are themselves regime-dependent?

We are only beginning to explore these questions. But one thing is clear: a regime-based approach forces us to confront markets as they are, not as we wish them to be. It allows us to move beyond static models and rigid classifications, adapting dynamically to an ever-evolving landscape. This paper is just the beginning, and I look forward to sharing further insights as our research progresses.

Download the full paper here.