Essays on microeconomic prediction problems : with applications to firm growth and public procurement

Economics and other social sciences have traditionally focused on developing and testing theoretically grounded causal explanations of individual and collective behavior. In contrast, this dissertation consists of four essays on microeconomic prediction problems, where the focus is on accurately predicting unobserved outcomes or outcomes that materialize in the future. The essays convey two complementary perspectives: economics can be helpful for prediction, and prediction can be helpful for economics. Across four distinct essays, this research studies prediction problems within two empirically rich contexts: high-growth enterprises and public procurement.

The first essay develops a tailored machine learning procedure to assist a budget-constrained venture capitalist in sourcing future high-growth enterprises. By optimizing for decision-relevant errors rather than standard statistical prediction errors, the method significantly improves the accuracy of predicting high-growth enterprises and reveals complementarities between algorithmic screening and human expertise.

The second essay leverages the predictive power of machine learning as a benchmark to assess the empirical adequacy of existing theories of firm growth. It applies a measure of completeness to quantify the share of predictable variation a theory explains, distinguishing this from irreducible noise that no theory is expected to capture. The results challenge persisting views of the adequacy of firm growth theories and provide directions for future theorizing.

The third essay evaluates a prospective entry-enhancing policy in public procurement, targeted at a subset of procurement auctions using ML-generated predictions of entry. This application combines methods from prediction and causal inference to optimize policy targeting and evaluates the intervention before implementation. Counterfactual simulations imply substantial potential cost savings by targeting the policy algorithmically to auctions predicted to receive few bids in the absence of the policy. Moreover, the simulations suggest that algorithmic policy targeting offers policymakers flexibility in choosing a suitable policy instrument in terms of effectiveness and implementation costs.

The final essay proposes and demonstrates a novel empirical strategy to detect favoritism as a mode of corruption in public procurement. By analyzing winning bid backlogs and bid submission timing through a regression discontinuity framework, the study uncovers patterns consistent with favoritism.

Collectively, these essays contribute to nascent, prediction-oriented approaches within microeconomics, a discipline that often overlooks prediction. By integrating ideas and theory from economics with advanced data-driven prediction methods, this dissertation delivers practical solutions for a variety of decision-makers, generates novel insights, and opens new avenues for future research.

Contact Joosua Virtanen

Email: joosua.virtanen@hanken.fi
LinkedIn: https://www.linkedin.com/in/joosuavirtanen/