Challenging Economic Dogma: Imperfect Knowledge, Unpredictability, and the Failure of Stationary Models

The podcast critically examines the foundations of modern economic forecasting and theory, arguing that the pervasive assumption of stationary distributions is fundamentally flawed. It highlights how "imperfect knowledge," particularly regarding unanticipated shifts in underlying economic processes, renders traditional models like Dynamic Stochastic General Equilibrium (DSGE) and the Rational Expectations Hypothesis (REH) invalid for real-world application. The speaker distinguishes between different levels of unpredictability—intrinsic (known knowns from chance), instance (known unknowns from fat-tailed distributions), and extrinsic (unknown unknowns from sudden distributional shifts)—emphasizing that the latter, characterized by location shifts, is the most problematic for current economic paradigms. These shifts, which are empirically common across various economic and social phenomena, lead to systematic forecast failures and undermine the ability to perform accurate intertemporal calculations. A core argument is that the five well-known theorems about conditional expectations, which underpin much of modern econometrics, only hold true under the unrealistic assumption that distributions never shift. When distributions do shift, these theorems fail dramatically, making the mathematical basis of DSGE models invalid and leading to "appallingly bad forecasts." The speaker illustrates this with numerous historical examples, from UK GDP growth and population changes to CO2 emissions and financial crises, demonstrating the constant non-stationarity of economic data. This constant shifting means that agents operating under rational expectations, as typically modeled, would appear "stupid" in the face of real-world changes, failing to adapt to new realities. To address these challenges, the podcast introduces advanced modeling tools such as "step indicator saturation" and "impulse indicator saturation." These methods are designed to detect and account for multiple location shifts and other non-linearities in data, even when dealing with a large number of variables and observations. The speaker asserts that these tools make it "almost costless" to check for breaks and shifts, allowing for empirical model discovery and theory evaluation that is robust to the ever-changing nature of economic systems. This approach moves beyond simply identifying forecast failure to providing a framework for building more empirically relevant and adaptive economic models. The broader implications are profound, calling for a fundamental re-conception of economic science. The current reliance on complete stochastic processes and probabilistic descriptions of change leads to misleading conclusions about rationality, market efficiency, and policy effectiveness. Instead, the podcast advocates for an approach that acknowledges and models adaptation to a constantly changing environment, much like NASA adjusts spacecraft trajectories in real-time. This shift in perspective is crucial not only for improving forecasting accuracy but also for developing more effective and appropriate policy responses that account for the inherent unpredictability and non-stationarity of the real world.

imperfect knowledge has profound consequences far beyond forecast failure in the real world

I've defined unpredictability as irreducible uncertainty

some events are so unpredictable that reasonable probabilities cannot be assigned

extrinsic unpredictability or unknown unknowns... occur when the entire distribution suddenly shifts

it wrecks the ability to do intertemporal calculations of the kind that dominate in modern macroeconomics

the five theorems about conditional expectations are true under the assumption that distributions never shift

economic theory suffers most and I mean most from distributional shifts

forecast failure does not imply either the theory is wrong or your forecasting algorithm is wrong it just implies something's happened you did not expect

unless you people he meant economists because I was an assistant professor solve the problem of what David would call location shifts and structural changes there'll be no finance to speak of that would be empirically relevant and we know macroeconomics is simply irrelevant

re age is the only rational way to forecast not the phrasing conditional on your assumption that the world can be described as a stochastic process adequately