Abstract: Taking the agricultural and forestry sector stocks of A-shares in the Shanghai and Shenzhen stock markets as research samples, with the period from 2021 to 2024 as the empirical research interval, the cointegration equation was used to select agricultural and forestry sector stocks with a common moving trend in the price series during the pairing period; and then, the stationarity test was conducted on the price difference series of stocks with cointegration relationship. For asset pairs that meet the stationarity requirements, their weekly realized volatility, realized skewness, and kurtosis are calculated, we used LSTM network to predict the residual convergence level of stock pairs in the short term, and finally simulated trading on eligible asset portfolios during the testing period, comparing the return level with the benchmark return. The paired trading asset portfolio of agricultural and forestry sectors constructed based on cointegration methods and LSTM networks can achieve excess returns while controlling risks.