["Journal of Risk and Insurance, EarlyView. ", "\nAbstract\nWe study U.S. property‐casualty loss reserve accruals, focusing on understanding why industry aggregate accruals exhibit large deviations from the losses that are ultimately paid. We start by comparing simple actuarial estimates based on individual company data to actual initial accruals and to actual ultimate loss reports. We find that individual company accruals track with actuarial estimates produced by Bayesian methodologies that smooth results over time, rather than methodologies that leverage current results. Although company accruals forecast ultimate results at the company level accurately, aggregated accruals perform poorly relative to aggregates from actuarial methodologies that leverage current results. These findings suggest that smoothing is individually rational but suppresses information that would otherwise contribute to the “wisdom of crowds” in aggregate, thereby distorting industry benchmarks used in pricing and reserving. More generally, our results offer a new explanation for why the informativeness of accounting aggregates deviates from that of their firm‐level counterparts.\n"]