The Pretence of Amnesia: Autocorrelation, Systemic Memory, and the Limits of Temporal Isolation

Standard econometric pedagogy and practice rest upon a foundational, yet highly fragile, epistemological assumption: that unobservable error terms are independent across time and space, expressed mathematically as �(u_iu_j) =0. By requiring the residuals of a model to exhibit zero covariance, this framework treats the continuous, dynamic flow of human economies as a sequence of isolated, memoryless events. However, because economic variables operate within deeply entangled, complex systems, omitted factors—such as institutional trust, cultural risk appetite, or market sentiment—naturally possess historical inertia. While the discipline frequently acknowledges the presence of systemic memory, its traditional remedies, such as artificially inflated standard errors (HAC) and autoregressive parameters (ARIMA), function primarily as mechanical patches that obscure deeper structural misspecifications. Furthermore, even highly sophisticated modern techniques, including Machine Learning causal forests and Synthetic Controls, fail to resolve the core epistemological crisis, as they remain bound by assumptions of total observability or historical stasis. We argue that true statistical rigor in complex, non-ergodic environments requires a paradigm shift. The discipline must pivot away from the hubristic pursuit of precise causal extraction and instead embrace quantitative methods designed to mathematically bound systemic exposure to path-dependent uncertainty