Faced with the traditional way of investing in gold and the new way of investing in Bitcoin, it is the dream of many people to develop a scientific investment strategy in order to get the maximum return. In this paper, we explore the maximum return an investor can get in five years with an initial capital of $1,000 and the effect of daily trading conditions and trading prices on the model, mainly based on a prediction model and a planned trading model. We begin with data preprocessing to analyze daily trading conditions and label whether gold can be bought or sold on a trading day. To improve the accuracy of the prediction model, we process the data using wavelet noise reduction techniques. Then linear regression prediction model, gray scale prediction model and recursive neural network prediction model are established to predict the next day’s data using the historical data before the trading day. Among them, the recursive neural network uses an improved version of the GRU model based on the LSTM model, which is faster in calculation. In order to make the prediction more reasonable and in line with the pattern of rest and relaxation, we make the prediction on a 30-day cycle. After a detailed comparison and analysis of the three models, we finally used the grayscale prediction model for short-term forecasting and the GRU model for long-term forecasting. We then built a planning trade model that puts itself in the shoes of a real investor and analyzes in detail all the strategies an investor can take on that day, coming to the conclusion that the investment situation can only be all cash or all bitcoin or all gold.