Considering the random factors involved in estimating modelfitting parameters, the classical linear regression model should be adjusted to improve its predictive performance. Therefore, this paper proposes an Lmultiplier model based on SignaltoNoise Ratio adjustment, which links the optimally fitted model to the best predictive model. The paper provides an explicit solution for the optimal Lmultiplier, and considering the presence of autocorrelation in the predictive variables, presents an improved expression for L. Subsequently, this paper discusses two estimation methods for the Lmultiplier and compares their respective advantages and disadvantages. Empirically, the theoretical validity of the Lmultiplier model has been confirmed by applying the modified Lmultiplier model to the problem of predicting stock returns. The results show that: 1)Compared to the baseline linear regression model, the model modified with the Lmultiplier has higher predictive accuracy; 2)When the noise is higher, the historical sample size is smaller, or the information content of the predictive variable is lower, the adjustment intensity of the Lmultiplier is greater, resulting in a more pronounced improvement in prediction; 3)For investors with meanvariance utility, this improvement in predictive performance can lead to an enhancement in investment returns. These findings also withstand a series of robust tests.