Models are abandoned for availability of alternatives, not merely false predictions

Evidence

Nor is the failure of an empirical prediction, even a central one, in itself enough to explain the abandonment of the Levy model. As we saw in chapter 2, the Capital Asset Pricing Model made, at least once Sharpe had the confidence to embrace it, a prediction that was manifestly false that all investors would hold portfolios of risky assets that were identical in their proportionate composition but that did not stop the model being adopted widely. What was at least as important as the empirical doubt cast on self-similarity or invariance under addition in explaining the abandonment of the hypothesis of Levy distributions was that a variety of other ways of modeling distributions with high frequencies of extreme events were becoming available (see, for example, Clark 1973). — Donald MacKenzie, An Engine, Not a Camera, 514

Categorization

• XREF: Directly echoes the performativity thesis: a manifestly false CAPM prediction did not prevent adoption, while Levy’s empirically-doubted model was dropped — the crucial variable is the availability of workable alternatives, linking to the book’s core argument about models shaping rather than merely describing markets. — connects to: Deployed machine-learning models often persist unchanged for decades in production; An influential historical model persisted for decades without direct evidence — Robert Irwin, For Lust of Knowing: The Orientalists…, 360; Belief systems reject disconfirming evidence rather than revising theories (+3 more)

Broader Connections

• Deployed machine-learning models often persist unchanged for decades in production #exemplifies (0.80)
• An influential historical model persisted for decades without direct evidence #exemplifies (0.75) — Robert Irwin, For Lust of Knowing, 360
• Empirical testing can only fail to disconfirm, never confirm, models like the CAPM #qualifies (0.72) — Donald MacKenzie, An Engine, Not a Camera, 408
• Belief systems reject disconfirming evidence rather than revising theories #exemplifies (0.70)