How does managerial perception of uncertainty affect corporate investment during the COVID-19 pandemic: A text mining approach

We propose a three-phase framework for developing sentence-level measures of uncertainty based on Japanese annual reports using the Bidirectional Encoder Representations from Transformers (BERT) model. This approach successfully distinguishes the uncertainty caused by the COVID-19 pandemic (COVIDU) from economic policy uncertainty (EPU). Furthermore, we control for firm-level business expectations (first moment effect) and examine the impact of managerial uncertainty perception (second moment effect) on corporate investment. We obtain the following two key results: First, our uncertainty perception measures strongly correlate with the proxy variable of dispersion in managerial sales forecasts, which is a frequently used measure of uncertainty. Second, corporate investment is strongly and robustly negatively related to higher uncertainty perception, with a 1-standard-deviation increase in sentences associated with uncertainty leading to a 3.57 % reduction in capital expenditure. We demonstrate the robustness of our results to the endogeneity problem by using instrumental variables. • Our uncertainty perception measures based on BERT model positively correlate with managerial forecast error. • Firm-level investment is negatively related to higher uncertainty perception. • Sentence-based uncertainty metrics exhibit explanatory and predictive power for capital investment. • We demonstrate the robustness of our results to the endogeneity problem by using instrumental variables.