ISE 537 Spring 2025, GPP
Final Project
Due Friday, May 9, 2025 by 11:59 pm.
Please select one of the three topics listed below for your final project. The final project should be completed individually. A guideline on the minimum components is listed under each topic. General instruction is provided at the end and it applies to all topics. Please submit your final report before the deadline and late submission is not allowed.
1. Dynamic Trading Strategy Design
• The auxiliary materials Lecture Notes 6-9 and Jupyter Notebooks (“Lecture 7-8 Regres- sion for Pair Trading.ipynb” are related to this project.
• Design an improved version of the pairs trading strategy (introduced in class) by includ- ing more stocks in your strategy.
• You can consider using the top 100 stocks (https://companiesmarketcap.com/) and their daily close price during the period of 2024 - 2025 [Note: you may not have access to all the stocks from Yahoo Finance during this period; please explain how you handle this issue].
• Explain whether dimension reduction techniques are needed.
• Explain how you measure the performance.
• Please show both the in-sample and out-of-sample performance.
• Show evidence that the strategy you develop is profitable.
2. Price Prediction: Comparison between ARIMA Model and LSTM
• The auxiliary materials Lecture Notes 10-12 and 16-17 and Jupyter Notebooks (“Lec 10-11 Time Series,” “Lec 12 Time Series.ipynb,” “Lecture 16 Deep Learning.ipynb,” and “Lecture 17 Stock Price Prediction using LSTM.ipynb”) are related to this project.
• Choose two stocks as you wish for the prediction. Do predictions separately for each stock.
• Pick a period (explain why) and use daily close price from Yahoo Finance for the pre- diction.
• Show evidence of convergence for the LSTM model and explain how you pick the lags for the ARIMA model.
• Compare the out-of-sample performance between ARIMA Model and LSTM. (Remember to do this in a rolling fashion!)
• Explain the pros and cons of both methods. Any difference between the performances of the two stocks you pick?
3. Reinforcement Learning for Optimal Execution
• The auxiliary materials Lecture Notes 13, 20-23 and the Jupyter Notebook “Lecture22- Reinforcement Learning for Optimal Execution.ipynb” are related to this project.
• Train the Q-learning algorithm with the simulation environment provided in the Jupyter Notebook (“Lecture 22: Reinforcement Learning for Optimal Execution”) on BrightSpace.
• Show evidence of convergence.
• Use Sharpe Ratio as the criterion and compare the performance of Q-learning to the following two strategies:
(a) executing with a constant trading speed, (b) executing everything at time 0.
• Design a new simulation environment and repeat the above procedures. Explain your results.
General Instruction for Final Projects:
• We would like you to on the one hand, approach the problem from a practical perspective, and on the other hand, have a systematic methodology. You are encouraged to go beyond the requirements mentioned and use the quantitative techniques you have to solve the problems.
• The final report is expected to contain the following components:
— Introduction to the dataset you use, the data format and the steps you perform for data pre-processing.
— Brief introduction to the models you use in the project.
— Detailed discussion for the model evaluation: proper choice of evaluation criterion (or loss function), out-of-sample test, comparison to benchmark models.
— Detailed explanation on how model parameters are selected.
— Detailed explanation of your results.
— Discussion on financial interpretation or economic insights of your model.
— A conclusion paragraph to summarize the methodology and your results.
— *Optional* If you explore some other related aspects for the project of your choice, extra credits (up to 20 points to the final grade) will be granted based on the quality of the additional materials. If you decide to do some extra work, please clearly indicate what is additional in your report.
• Do not include your codes in the report (nor attach them at the end). Codes should be submitted via a separate submission channel on Brightspace.
• Page limit: minimum of 3 pages and maximum of 10 pages.
• You can use the Python codes on Brightspace or any open-sourced codes you find online. Please acknowledge the source in your report if you use open-sourced codes.