We are using a set of classifiers, regressors together with depp LSTM network for market prediction. The following techniques are used:

• Logistic regression (binary, multinomial)
• Gaussian naive Bayes
• Support vector machines
• k-nearest neighbors
• Decision trees
• bagging and boosting techniques (AdaBoost)
• artificial neural networks
• deep LSTM networks

We are using the following technical indicators to extract data features:

• RSI
• Stochastic K%D
• MACD
• CCI
• ATR
• ADL
• Williams R
• price rate of change
• on balance volume

We create the following response (endogenous) variables:

• binary classification (up/down)
• tertiary classification (up/sideways/down)
• multinomial classification (strong up/up/sideways/down/strong down)

The tertiary and multinomial classification is relative to the volatility where we use EWMA volatility and drift-independent volatility (Yang & Zhang 2000).

Input files are expected to be csv files according to the following format:

The first lines are header lines, which contain the following key: value pairs:

• Symbol: stock symbol
• Name: company name
• Exchange: stock exchange

The following line is the header for the csv columns.

The rest of the file is the market data. The file is expected to contain the following fields: date, open, high, low, close, volume.

To amend an input file with the data from the technical indicators, run

python gen_indicators -o <output-file> <input-file>

To run all the classifiers, regressors and deep networks on an input file, run

python fit_models.py <input-file>