Beyond Static DCF: Machine Learning-Enhanced Free Cash Flow Forecasting and Equity Valuation in Nepal’s Hydropower Sector

Abstract

This study develops a machine-learning-enhanced framework for Free Cash Flow to Firm (FCFF) forecasting and intrinsic equity valuation in Nepal’s hydropower sector, where data scarcity and operational volatility limit static forecasting assumptions. Using panel data for 105 Nepalese hydropower companies covering 2010–2025, the study evaluates Support Vector Regression (SVR) against conventional and alternative machine-learning approaches. GroupKFold cross-validation reduces firm-level data leakage, while permutation importance and partial dependence analysis provide interpretability. FCFF forecasts are incorporated into a discounted cash flow (DCF) framework and compared with a zero-growth baseline. The tuned SVR model achieved a Mean Absolute Error (MAE) of 15.8%, compared with 20.8% for the baseline, showing a 24% improvement in forecasting accuracy. In absolute currency, average errors were Rs. 91,442 for SVR and Rs. 96,114 for the baseline. The Wilcoxon test indicated consistent directional improvement, while Permutation Importance identified ROA, Working Capital, and Installed Capacity as the main predictors. In DCF valuation, SVR-derived intrinsic values showed a ~4.01% reduction in average valuation error compared with static baselines (IQR: Rs. 2.37M–Rs. 5.86M vs. Rs. 2.42M–Rs. 6.22M). Keywords: machine learning, free cash flow, intrinsic valuation, support vector regression, hydropower, Nepal stock exchange