This section details how a user can utilize the predictive system I have created for stock trading. It should be noted that the user must have access to Excel and Python.
1. ACCESSING THE CODE
The complete code for my finished project is available on GitHub (https://github.com/lhladik15/Capstone499). The user should be able to fork the repository for the folder labeed “Capstone499.” Forking allows the user to get a copy of and freely experiment with the code.
2. SETTING UP YOUR INPUTS
The inputs to the system should be contained within one or CSV files. (Examples of such CSV files are kept as referrals in my GitHub repository). The data should be formatted such that there is a date column indexing the data, and then daily data for numerical data relating to stock price movements. This could span from historical stock prices to technical indicators to fundamentals to whatever the user deems appropriate data corresponding to stock prices.
In line 47 in the code, there is a variable called filesnames, which will store the file names of each of the input CSV file(s). The code on GitHub is as follows:
filesnames = glob(‘IT1_*.csv’)
The name’IT1_*.csv’ in quotations corresponds to the names of the input CSV files. For my tests, I was utilizing IT sector stocks, so I began each file name for each stock with the letter ‘IT’ to denote this. This notation with the asterisk (*) means that the filenames of every stock starting with the letters ‘IT’ will be read in for later calculations.
3. SETTING THE IDEAL PARAMETERS
In accordance with my design requirements, I made the final design output flexible to the user. Thus, there are a number of configurable parameters. The number of days to forecast out on for stock prediction is represented by the variables ‘numForecastDays’ and can be set to any desired number in line 68. The number of lag days for a lagging window (represented by the variable named ‘forecast_out’ can be set in line 76). Any number or type of numerical input can be put into the code. The names of these variables that correspond to the CSV file must be specified in line 66 for the variable ‘target1’ and in line 253 for the variables ‘df_InvertedVals’. Lastly, the code is set up so that it will output two tables—one containing the stock prices that are projected to rise the most and the stock prices that are projected to fall the most. The number of stocks desired in each of these tables can be specified by setting the variable ‘stockRef’ in line 328.
4. Getting Outputs
There two different outputs from the code. The first is the stock table mentioned above that projects which the topmost and bottom-most stocks. The second output is a CSV file or multiple CSV files. The number of these files corresponds to the number of inputs CSV files for stock that is being analyzed. In the CSV file the predicted price change, the actual predicted price change, and direction that the stock is projected to move are all output. Additionally, since I was strictly working with historical data to back-test the effectiveness of different algorithms, I knew the true prices that corresponded to the predictions. As such, these CSVs will also contain the true price, error between the true price and predicted price, the true price movement, and the actual direction the true price moved. Again, because this projected involved back-testing, the cost metrics developed were also incorporated into this CSV file output. One output will be for the profits the investor would have made if he had invested based on the algorithms predictions. For the sake of comparison, the market baseline (i.e., how well the investor would have done if he had just bought and held the stock) is also included to show if using the predictive algorithm would both minimize losses and maximize gains. (The ultimate goal is to use predictions to defeat the market baseline).
