Determinants of corporate cash holdings: An application of a robust variable selection technique

US firm cash holdings have become increasingly concentrated over time withering shareholder returns and heightening agency problems associated with free cash flows. We investigate the determinants of corporate cash holdings using a robust regression technique (the Least Absolute Deviation Regression, LAD) and a state-of-the-art variable selection procedure (The Least Absolute Shrinkage and Selection Operator, LASSO). This framework identifies a sparse model which is also resistant to outliers. We obtain several results. First, the median absolute errors of the LAD-LASSOselected variables is significantly lower than the corresponding values of the basic determinants commonly used in the literature, in both in-sample and out-of-sample schemes. Second, financial leverage is a key determinant of cash holdings, indicating that cash and debt policies are tightly related (debt-cash substitutability). Third, none of the corporate governance proxies identified in the literature as a corporate cash holding determinant (managerial ownership, board independence, CEO duality, board size, and institutional ownership) is selected by the LAD-LASSO method, challenging the potency of governance mechanism in corporate cash management. Fourth, the financial crisis of 2008 dramatically changed the selected predictors for cash holdings as well as requiring additional predictors for the period, though their number was heavily curtailed after the period. Fifth, the money demand function is different between financially constrained and unconstrained firms, in both combination and number of influential variables. Finally, prediction error of the LAD-LASSO-based cash model is smaller than that of the principal components regression in both in-sample and out-ofsample prediction. Our findings are of importance to corporate cash managers, shareholders, monetary policy authorities, and scholars on the issues such as adjustment speed of cash holdings and valuation of excess cash reserves, which are based on the fitted values from the cash model.