A question of time
I’ve been thinking about time recently. In economics and finance, time is central to our analysis. Most of the information we care about, after all, is spread and dispersed across time. As a result, the way we embed our models with information, and how we sort and organize that information, is often a question of time. The information embedded in time is powerful, often overwhelming.
Imagine you wanted to predict the price of the S&P 500 by tomorrow’s close or the level of GDP in the next quarter. You could do a lot worse than looking at the value from the previous period and applying a multiple, often close to one. This simple heuristic is, in fact, a powerful tool in time series forecasting. For those versed in time series econometrics, this is recognized as an autoregressive process, specifically AR(1). In plain language, this means that the information contained in yesterday’s value of a time series is often a good predictor of its value tomorrow. This is the case for a whole host of economic and financial time series data.
This perspective changes when you manipulate your data to remove unit roots or trends. In theory, differencing the data should eliminate persistence. In practice, however, even time series in first differences often exhibit significant autocorrelation. The shadow of the past weighs heavily, even on supposedly well-behaved data. Brute force can always be applied to produce mean-reverting data, but the question remains whether such data bear any resemblance to the original series of interest. Often, they do not. Empirical evidence suggests that it is difficult for forecasters to devise a model that consistently outperforms the predictive power of the historical data itself. History, it seems, sets a high bar for predicting the sudden, large swings in events and data that move markets and economies. This is as much a philosophical question for researchers as it is a practical problem—one often addressed through statistical alchemy in the search for patterns in historical data that might hold predictive power.
Macroeconomics has maintained a somewhat narrow conception of time since the rational expectations revolution, which followed the famous Lucas Critique, and paved the way for the neoclassical models that dominate contemporary literature. But reality is often more complex than what economic models portray. In, “The Problem of Time in the DSGE Model and the Post‑Walrasian Alternative,” Perry Mehrling contrasts the classical economic view of time with a newer, so-called Post-Walrasian perspective. He begins by interrogating the dominant forward-looking conception of time that underpins Dynamic Stochastic General Equilibrium (DSGE) models, the backbone of modern macroeconomic theory. In these models, the engine of economic behavior is anticipation: rational agents forecast future conditions and plan accordingly, optimizing intertemporally under uncertainty. Here, the present is shaped by expectations.
In contrast, Mehrling draws inspiration from classical economic traditions by emphasizing an alternative, past-determined view of time. In this view, current economic realities emerge from entrenched institutions, historical path dependencies, structural inertia, and the legacies of prior economic regimes. The present, in effect, is not an open slate but a canvas shaped by what has already occurred, a reality embedded in social norms, financial structures, and institutional arrangements that evolve slowly over time.
Mehrling sees these perspectives as complementary. He proposes a Post‑Walrasian synthesis that integrates both: the present is shaped by the dual forces of historical legacy and forward-looking anticipation. As elaborated in the broader Post‑Walrasian literature, this synthetic view recognizes that agents operate under conditions of bounded rationality and limited information, which preclude the kind of pure, forward-looking optimization assumed in DSGE models. At the same time, it acknowledges that institutions—whether financial, regulatory, or cultural—imbue the present with inertia and context, both constraining and enabling economic behavior.
According to Mehrling, this synthesis addresses several key critiques of conventional macroeconomic modeling. While DSGE frameworks offer mathematical elegance and clear policy prescriptions, they often fall short in accounting for real-world phenomena such as institutional stickiness, coordination failures, endogenous instability, and sudden crises. A Post-Walrasian outlook, by contrast, is better suited to accommodate these dynamics. It imbues time with both memory and anticipation, situating economic agents within evolving, path-dependent institutions while also recognizing their efforts to plan for an uncertain future. In the Post-Walrasian world, the present is not merely a forward arrow or a backward-looking shadow, but a dynamic intersection of paths—where history and hope converge to shape economic evolution.
The question, in a nutshell, is this: how much information do agents have access to, how do they use it, and how useful is that information for predicting the future and making decisions under uncertainty? Neoclassical economics assumes that agents are highly capable of making such decisions, or at the very least, that economic models should proceed on that assumption. The Post-Walrasian critique argues that agents’ decisions are more nuanced, and that heterogeneity among agents introduces biases that cause different responses to similar objective information.
Mehrling’s discussion of time intersects subtly with the introduction of the simple autoregressive model. In such a model, agents have access to the full history of the data they’re trying to predict and make a "rational" decision about their future behavior based on that information. In an AR(1) model, this typically means assuming that the immediate future will resemble the recent past. The key question for researchers, then, is under what conditions yesterday’s value is not a good predictor of tomorrow, and whether such conditions can be identified in advance.
Historical volatility analysis—modeling the second, third, and fourth moments of a distribution—is one approach. But this often tells you the likely magnitude of forecast error, not whether your point forecast will be correct. This kind of insight is useful for many questions in economics and finance, for example the pricing financial instruments such as options or the underlying behaviour of a given data set, but it does little to improve the accuracy of the forecast itself.
The treatment of time in economic theory and forecasting reveals both the power and limitations of our models. Whether viewed through the lens of autoregressive processes or within the broader philosophical debate between forward-looking rationality and historically rooted inertia, time remains an inescapable and often elusive dimension of economic analysis. Models like DSGE provide valuable frameworks, yet they also expose the inherent challenges of prediction in systems shaped by uncertainty, structural complexity, and bounded rationality. As Perry Mehrling and the Post-Walrasian tradition remind us, understanding the present—and anticipating the future—requires not only technical sophistication but also a conceptual openness to time’s dual nature: as both a repository of historical legacies and a space of forward-looking expectations. Recognizing this duality may not make forecasting easier, but it does make our theories more honest—, and perhaps more useful in navigating the real-world dynamics.