Training Contracts, Worker Overconfidence, and the Provision of Firm-Sponsored General Training

Training by firms is a central means by which workers accumulate human capital, yet firms may be reluctant to provide general training if workers can quit and use their gained skills elsewhere. ``Training contracts" that impose a penalty for premature quitting can help alleviate this inefficiency. Using two exogenously staggered contract changes at a leading trucking firm, we show that training contracts significantly reduce post-training quitting and increase the profitability of training. We demonstrate further that training contracts have especially large effects on training profits in our setting because of their interaction with worker overconfidence. If workers are overconfident about their own productivity at the current firm relative to their outside option, they will be more likely to sign training contracts and less likely to quit after training. Combining weekly productivity data with weekly productivity beliefs, we show that workers systematically overpredict their productivity, both with and without randomized financial incentives for accurate prediction. To quantify the impact of overconfidence for optimal contracts and welfare, we develop and estimate a structural learning model with biased beliefs that accounts for many key features of the data. Eliminating worker overconfidence would moderately increase worker welfare by 1.7%, but would decrease training profits by over $8,000 per truck and substantially alter the optimal training contract. We confirm the feasibility of reducing overconfidence using an information experiment at a second large trucking firm. Despite the positive effect of training contracts on profits, training may not be profitable unless workers are overconfident.