Skip to main content

Please enter a keyword and click the arrow to search the site

Goodhart’s Law and Machine Learning: A Structural Perspective


International Economic Review



Authors / Editors

Hennessy C;Goodhart C

Publication Year



We develop a structural framework illustrating how penalized regression algorithms affect Goodhart bias when training data is clean but covariates are manipulated at cost by future agents facing prediction models. With quadratic manipulation costs, bias is proportional to sum-of-squared slopes, micro-founding Ridge. Lasso is micro-founded in the limit under increasingly steep cost functions. However, standard penalization is inappropriate if costs depend upon percentage rather than absolute manipulation. Nevertheless, with known costs of either form, the following algorithm is proven manipulation-proof: Within training data, evaluate candidate coefficient vectors at their respective incentive-compatible manipulation configuration. Moreover, we obtain analytical expressions for the resulting coefficient adjustments: slopes (intercept) shift downward if costs depend upon percentage (absolute) manipulation. Statisticians ignoring agent borne manipulation costs select socially suboptimal penalization, resulting in socially excessive, and futile, manipulation. Model averaging, especially over Lasso or ensemble estimators, reduces manipulation costs significantly. Standard cross-validation fails to detect Goodhart bias.

Available on ECCH


Select up to 4 programmes to compare

Select one more to compare
subscribe_image_desktop 5949B9BFE33243D782D1C7A17E3345D0

Sign up to receive our latest news and business thinking direct to your inbox


Sign up to receive our latest course information and business thinking

Leave your details above if you would like to receive emails containing the latest thought leadership, invitations to events and news about courses that could enhance your career. If you would prefer not to receive our emails, you can still access the case study by clicking the button below. You can opt-out of receiving our emails at any time by visiting: or by unsubscribing through the link provided in our emails. View our Privacy Policy for more information on your rights.