SPX_Options_ANN

Public repository for the code of my Master's thesis. My e-mail is: [email protected]. Contact me if you need any help or have a question.

The work is related to the paper "Option Pricing with Deep Learning" by Ke and Yang (2019). My code is inspired by what the authors made publicly available: https://github.com/ycm/cs230-proj. The objective is to price/evaluate european vanilla options (financial derivatives) using different kinds of neural networks.

Instructions

process_raw_data_phase1.py, process_raw_data_phase2.py and process_raw_data_phase3_final.py were used with the full proprietary dataset. process_raw_free_dataset.py is to be used with the small free dataset.

1. Run process_raw_data_phase1.py, process_raw_data_phase2.py and process_raw_data_phase3_final.py sequentially to create options-df.csv and underlying_df.csv in the "Processed data" folder. It uses data in the "Raw data" folder.

Or

1. Run process_raw_free_dataset.py to create options-df.csv and underlying_df.csv in the "Processed data" folder. It uses data in the "Raw data" folder. Because some data files are quite large I don't include them in the repository. The sources of the free raw data are:

• Underlying - https://www.wsj.com/market-data/quotes/index/SPX/historical-prices
• Options (September 2018 and August 2019, respectively) - https://www.historicaloptiondata.com/files/Batch_PRO_Sample_PHC.zip and https://www.historicaloptiondata.com/files/Sample_L2_2019_August.zip
• Treasury Rates - https://www.treasury.gov/resource-center/data-chart-center/interest-rates/pages/TextView.aspx?data=yieldYear&year=2019

2. MLP1 model (what is explained here also applies to other models):

   • Run mlp1_call.py and mlp1_put.py. Uses the data in the "Processed data" folder.
   • Run mlp1_error_metrics.py. It returns train-MSE (the mean squared error on the training set) and metrics on the test set: MSE, Bias (the median percent error), AAPE (the average absolute percent error), MAPE (the median absolute percent error), and PEX% (the percentage of observations within ±X% of the actual price).

3. MLP2 model:

   • Run mlp2_call.py and mlp2_put.py
   • Run mlp2_error_metrics.py

4. LSTM model:

   • Run lstm.py
   • Run lstm_error_metrics.py

5. Black-Scholes-Merton (BSM) model:

   • Run bsm.py
   • Run bsm_error_metrics.py

6. Create graphs showing training and validation losses over epoches - run ann_loss_graphs.py

Other

There is also code for a model called MLP3. It was supposed to be an MLP1-like model that outputed arbitrage-free option prices. In the end due to lack of time and expertize I couldn't fully implement it but I leave the code here. It was based on the work of Andrey Itkin (2019).