Why Radical Uncertainty Undermines Predictive Ambitions in Economics
By Elias Sanchez
In antiquity, kings, generals, and heroes journeyed to the Oracle of Delphi seeking guidance about the future. Their questions were strikingly similar to those posed by economists today: What will happen if I choose this path? Will victory come next season? Will prosperity return? The oracle, seated above fissures of intoxicating vapours, never gave precise forecasts. Her answers were elliptical, conditional, and interpretive. “If Croesus goes to war with Persia, he will destroy a mighty empire,” she told Croesus—but she did not specify whose. The message was merely a signal, not an itinerary blueprint.
In this sense, the Delphic pilgrimage mirrors contemporary economics. Today, policymakers, investors, and analysts consult complex models rather than priestesses, but they often expect the same prophetic clarity. They ask: What will GDP be next quarter? When will the bubble burst? Will the currency collapse next year? And like the oracle, economic models can offer only broad warnings—“markets appear fragile,” “credit conditions are tightening,” “imbalances are accumulating.” They cannot decree when or how events will unfold. The epistemic structure is the same: models, like oracles, speak in conditional tendencies, not deterministic chronologies.
This critique speaks directly to a recent article written by The Economist, on “Why Wall Street won’t see the next crash coming”. Financial actors rely heavily on autoregressive models, treating them as mechanical predictors rather than interpretive tools. Unlike the ancient Greeks, they overlook that all forecasts are embedded in uncertainty. Institutional pressures and herd behaviour amplify this error, transforming inherently probabilistic outputs into seemingly deterministic insights—and obscuring the very fragilities they aim to detect. Autoregressive models are induction formalised mathematically. They extend patterns from the past into the future, assuming stability that human action and radical uncertainty often undermine. The danger, in both antiquity and modern finance, emerges when warnings are mistaken for certainties.
The story of Croesus offers a paradigmatic warning for economic practice. Croesus misinterpreted Delphi’s prophecy and engineered his own downfall. Faced with uncertainty, he interpreted the Delphic oracle’s deterministic statement—“a great empire will fall”—as a deterministic prediction guaranteeing victory over Persia. The failure lay not in the oracle’s ambiguity but in Croesus’s desire for epistemic certainty where none could be found. He transformed a signal into a prophecy and suffered the consequences. Contemporary policymakers and financial actors risk a similar error when they treat uncertainty as a variable that can be fully quantified, modelled, and statistically neutralised. In doing so, they elevate model-based signals into mechanical predictions.
As economist Dimitri Zenghelis cautions, economic models are valuable analytical tools, but their capacity is limited: they can illustrate short-run ‘what-if’ scenarios, yet cannot reliably forecast long-term outcomes in systems shaped by innovation, learning, and end. The lesson is epistemological: models, like oracles, illuminate structures, not futures. They can warn—but they can never tell what will happen next Tuesday.
2. How Delphi Illustrates the Nature of Economic Forecasting
The Delphic analogy becomes analytically useful when we recognise that, just as ancient heroes sought prophetic clarity about an uncertain future, modern economic “heroes”—Nobel laureates, market analysts, renowned academics, institutional forecasters—also approach their own oracles in search of foresight.
The tools have changed, but the epistemic aspiration remains the same. Instead of priestesses, they now consult sophisticated econometric models, AI-driven predictive systems, and statistical frameworks that promise insight into the future.
Yet the Oracle of Delphi never delivered forecasts in precise chronologies. Her pronouncements were conditional, elliptical, and fundamentally interpretive. Modern forecasting, despite its veneer of technical rigour, mirrors this structure. Autoregressive models, VARs, DSGEs, and machine-learning predictors operate by extracting patterns from past data to infer tendencies about future developments. These techniques yield conditional probabilistic relations rather than deterministic outcomes, and they depend on the assumption that historical regularities will persist in a dynamic context characterised by non-ergodic factors such as novelty, regime shifts, and radical uncertainty.
The case of Croesus illustrates the perennial danger of mistaking conditional signals for definitive predictions. When told that “a great empire will fall,” he assumed the prophecy guaranteed his victory over Persia. What he failed to understand—much like those who overinterpret econometric outputs as iron laws—is that the message did not resolve the underlying uncertainty. His decision problem was far more complex than a binary “will or will not prosper” outcome, just as economic phenomena cannot be reduced to a joint-probability event in a closed statistical system.
The structural similarity becomes clear: both ancient divination and modern modelling offer signals, not certainties. They illuminate possible directions, highlight vulnerabilities, and reveal patterns of coordination or fragility—but they cannot specify the precise unfolding of future events. The Delphic example thus serves as a conceptual bridge, showing how predictive ambition collides with the inherent openness and uncertainty of non-ergodic human action.
3. From Oracles to Models: The Modern Parallels
The contemporary turn to economic modelling closely resembles the ancient reliance on the Oracle of Delphi. Instead of embarking on pilgrimages, today’s economists, investors, central bankers, and governments consult statistical, econometric, and increasingly AI-driven models in search of clarity about an uncertain future. Amid an AI boom frequently compared in scale to the dot-com bubble—one that some analysts speculate could erase trillions in market value if momentum reverses—models have become an indispensable tool for narrowing, or appearing to narrow, the gap between uncertainty and certainty.
The questions they pose are structurally identical to those once asked of Delphi: When will the turning point arrive? How will events unfold? Will this scenario happen at all? Forecasters use inductive methods—extrapolating from patterns in historical data—to extract signals that can guide expectations in the present. This inductive reliance on past regularities is not merely technical; it forms part of the trust architecture sustaining contemporary financial and policy decision-making. In global markets, expectations are signals, and signals require interpretation.
But just as Delphi never delivered specific chronologies, modern models do not output deterministic predictions. Their results come in the form of tendencies: fragility, imbalances, over-heating, credit tightness. In econometrics, these tendencies are expressed as point estimates or probability distributions derived from autoregressive structures, inference techniques, and hypothesis-driven assumptions. The technical details—stationarity constraints, identification strategies, model selection criteria—ultimately rest on simplifying assumptions about a social world characterised by heterogeneous agents, uncertain expectations, and continuous structural change, features of a non-ergodic system.
This reliance on simplified representations echoes the critique developed by F. A. Hayek in Law, Legislation and Liberty. Hayek warned against the Cartesian impulse to impose artificially precise models onto complex social processes. Such impulse labelled as constructivist rationalism treats economies as if they were mechanical systems rather than evolving orders shaped by non-ergodic human action and entrepreneurial discovery. In these open systems, dynamic forces—including Schumpeterian “creative destruction”—continuously alter the very structures that models attempt to quantify.
There is, however, a crucial distinction between using models as mere minor inputs for decision-making and treating them as authoritative determinants of policy. The former is compatible with an Austrian epistemology; the latter is not. When state actors elevate econometric forecasts into policy constraints—as in Rachel Reeves’s recent commitment to adjust the United Kingdom’s fiscal “headroom” based on the Office for Budget Responsibility’s growth projections—the inductive limits of modelling become politically consequential. Here, forecasts cease to be interpretive signals and risk becoming prescriptive commands.
This point aligns directly with The Economist’s observation that while models can say “markets look fragile,” they cannot say “the crash will occur next Tuesday.” The proper epistemic status of models—particularly when used in finance and public policy—is therefore Delphic: interpretive, suggestive, and fundamentally non-deterministic.
4. Austrian Epistemology: Why Predictive Precision Is Impossible
The methodological core of the Austrian critique lies in its rejection of induction as a foundation for economic knowledge. As Hans-Hermann Hoppe argues in his work The Austrian Method, inductive modeling—however statistically sophisticated—is deficient. It is bound to a Delphic conditionality because it extracts patterns from past data without addressing the non-ergodic essence that drives economies: human action itself. Its claims are interpretive rather than conclusive. From an endogenous perspective, contemporary inductivism further neglects the “butterfly effect,” whereby minute differences in initial conditions can generate radically divergent outcomes. This sensitivity undermines the reliability of simulations that assume ceteris paribus stability, overlooking the myriad pathways that economies may follow under different time, historical, institutional, and spatial configurations.
By contrast, Austrian epistemology is grounded in praxeology and deduction. Economic knowledge begins from the axiom of purposeful action: individuals employ scarce means to pursue subjectively chosen ends. Prices, marginal valuations, and market outcomes do not emerge from statistical regularities; they emerge from the deliberate choices of agents confronting scarcity. From this starting point, Austrian theory deduces the logic of exchange, the structure of production, and the formation of prices. Marginal value, for instance, is understood a priori as a subjective ranking of ends, not as a statistical artefact extracted from observed data.
This deductive foundation reveals why economic understanding must proceed from the logic of human action upward, and not from data downward. Empirical and econometric observations can illustrate theory but cannot logically ground it. A reconciliation is possible—but only in one direction: deduction must come first. Only once the causal structure of economic phenomena is understood theoretically can inductive tools be used, in a Delphic and modest way, to provide auxiliary signals or partial guidance.
The importance of this distinction becomes clear in the context of radical uncertainty. The future in a capitalist economy is shaped by non-ergodic entrepreneurial creativity, competitive disruption, and continuous innovation—forces that cannot be compressed into stochastic models. Human action is open-ended, non-repetitive, and embedded in shifting institutional frameworks. For this reason, the future cannot be known probabilistically in the same way that physical systems can. Radical uncertainty is not a temporary technical limitation; it is an inherent feature of social reality.
Thus, attempts by financial institutions and policymakers to decode the future through increasingly complex models—including machine learning systems and AI-driven predictive architectures—encounter this limitation directly. No amount of computational power can eliminate the fact that models rely on past data, while economic reality is driven by future-oriented, imaginative action that often breaks with historical patterns.
Thus, the Delphic analogy is not merely poetic. It exposes a principled epistemological boundary: economic models, like ancient oracles, can signal vulnerabilities, tensions, or emerging fragilities, but they cannot deliver deterministic forecasts. The logic of human action itself rules out predictive precision.
5. The Risk of Misinterpretation: Croesus and Modern AI Policymakers
Similar to the case of Croesus, policymakers and economists often risk committing the same error. Faced with complex financial systems and political pressures, they may overinterpret the outputs of models, treating them not as structural tendencies but as precise forecasts. The lead-up to the 2008 financial crisis is an instructive case: stochastic models, Value-at-Risk frameworks, and stress-testing architectures failed to anticipate the magnitude and nature of the collapse. Yet their outputs were treated as authoritative, encouraging the belief that tail risks had been tamed and markets were understood.
The current wave of AI-driven predictive enthusiasm exhibits similar patterns. Machine-learning systems promise unprecedented forecasting power, yet they remain fundamentally inductive, trained on historical data and unable to anticipate structural breaks, entrepreneurial innovations, or shifts in regulatory and institutional frameworks. Their outputs may appear more sophisticated than earlier stochastic models, but their epistemic limitations are the same: they offer patterns, not prophecies.
This misinterpretation becomes particularly dangerous in the realm of public policy. Governments that base fiscal, monetary, or regulatory decisions on model-driven forecasts risk turning interpretive signals into prescriptive rules. When policymakers assume that a model’s projection of growth, inflation, or debt sustainability reflects a deterministic future, they commit the Croesian error: they treat conditional knowledge as certain. The illusion of precision leads to overconfidence, rigidity, and policy mistakes.
The practical lesson is therefore clear. Economic models—whether econometric, stochastic, or AI-based—should be read as structural indicators, not deterministic predictive oracles. They highlight vulnerabilities, map potential trajectories, and help identify coordination failures or imbalances. But they cannot pinpoint outcomes in a capitalist economic world shaped by radical uncertainty, entrepreneurial creativity, and institutional change.
Like Croesus, decision-makers who mistake signals for certainties risk disaster. The Delphic analogy thus reveals the enduring danger of conflating interpretive tools with prophetic authority, reminding us that economic foresight is always—and must remain—bounded and conditional.
6. Conclusion: What Models Can Do
The Delphic analogy clarifies the proper epistemic status of economic modelling. From an Austrian perspective, models have genuine analytical value, but only within strict boundaries. They can illuminate structural patterns, map qualitative tendencies, and draw attention to fragilities, discoordination, or emerging institutional distortions. In this sense, models contribute to understanding the contexts in which economic processes unfold. They inform judgement rather than replace it.
What they cannot do is provide precise chronologies or deterministic forecasts. No model—whether econometric, stochastic, or AI-driven—can specify the timing, magnitude, or causal sequence of future events. Human action, as a force that move economies, is purposive, open-ended, and driven by dispersed knowledge and entrepreneurial creativity. These features make the social world fundamentally non-ergodic. The future is not an extension of the past, and its key determinants cannot be reduced to stable statistical parameters.
Thus, the role of models is necessarily Delphic. They offer clues rather than certainties. They speak in conditional signals—“markets look fragile,” “debt dynamics are becoming unsustainable,” “credit conditions are tightening”—but they remain silent on the precise moment or form that future events will take. Their guidance must be interpreted, not obeyed; integrated into judgement, not substituted for it.
If Croesus erred by mistaking a warning for a prophecy, modern policymakers risk repeating the same mistake whenever they treat model outputs as fate rather than as fallible indicators. The Austrian lesson is that economic foresight is inseparably bounded by radical uncertainty. Models assist understanding, but they cannot eliminate the unpredictability inherent in human action. In this sense, the title’s claim is vindicated: models warn, but never predict.
