Corporate Governance and Equity Evaluation: Nonlinear Modeling via Neural Networks

Abstract

This study queries the linear information dynamics (LIM) assumption of the Ohlson \r\n(1995) valuation model, for it is as if the assumption of linear information dynamics (LIM) \r\ndoes not exist. Prior studies used the OLS model to estimate the relationship between firm \r\nvalue and corporate governance but in the wrong way. This may have been due to problems \r\nwith the model’s specifications which led to the wrong empirical results. Thus, the purpose \r\nof this paper is to demonstrate that the artificial neural network (ANN) model is better than \r\nthe OLS model. Moreover, we will examine whether a nonlinear model created by an \r\nartificial neural network (ANN) model will perform the best in predicting firm value. \r\nThe empirical results indicate that the proposed neural network model can forecast \r\nfirm values more accurately and have better explanatory power than the conventional OLS \r\nmodel. Even after 100 epochs of iterative simulation, the neural network still outperforms \r\nthe OLS model in terms of explaining the training sample, verification sample, testing \r\nsample, and the holdout sample, with the confidence levels ranging from 99%~100%. The \r\nforecasted results are also tested using differential analysis. It is discovered that the MSE is \r\nextremely low, meaning that the accuracy of the neural network model is very high. The \r\n100-epoch simulation and sensitivity test both empirically validate the robustness of the \r\nresearch results. The superior forecasting capability of neural networks found in this paper \r\ncan be a reference for both the regulator concerned and for investors in decision making.