A Quantitative Way To Forecast Company Fundamntals:the Lookahead Factor Model

In the stock market,investors will adopt a variety of investment strategies.But in recent years,value investment team has grown larger.And the overall investment style in the A-share market is also changing from speculation and technical analysis to investment and fundamental analysis.Quantitative investment is also a fast-growing investment method in China.Factor investment,which is a popular kind of quantitative investment,is based on factors designed with fundamental data.However,traditional factor models aim to predict stock returns directly,while factor investment strategies simply use published data.Different from the traditional factor investment,our product attempts to forecast companies' fundamentals to obtain a more stable model,then uses the result of the model as forward factors for factor investment.The fundamentals-forecasting model trained by mass of data is the core of our product,which called the Lookahead Factor Model(LFM).Specifically,after selecting the variables,while obtaining and preprocessing the data,the data is divided into two sets,which are the in-sample training set and the out-of-sample testing set.Then,among several models,this article chooses LSTM neural networks as the structure of LFM,the lookahead factor model,and performes a 10-fold cross-validation.The results show that LSTM can adapt to the task.When backtesting on testing set,we obtain the trading signals for each trading day,based on the predicted results of LFM outside the sample.With 25 stocks in full positions and equal weights,the cumulative yield of 44.6% and annualized yield of 13.76% were obtained during October 31 st 2016 to September30 th 2019,far exceeding the traditional single-factor strategy benchmarked by this strategy.Compared with the performance benchmark HS300,the fluctuation of the yield of this strategy is slightly larger.However,this exchanges for a high excess return------Sharpe Ratio and Sortino Ratio are satisfactory,clearly outperforms the benchmark.Finally,the attribution analysis using the backtest results of the strategy is performed.From the results of Brinson's attribution analysis,since our strategy has kept stocks in full position,the impact of active allocation in excess returns is small.The excess returns mainly come from the stock selection.From the factor analysis perspective,both the FF five-factor analysis and the Barra factor analysis show that this strategy prefers to allocate small-cap growth stocks,instead of large-cap stocks and value stocks.All in all,our product outperforms both the traditional factor investing and HS300,and this article shows its style and source of return by assessing and attribution analysis,to assure its reliability.Besides,this article extends the use of fundamentals-forecasting model,and provides a reference for data analysis and factor investment.