代写FINE 695 Applications of AI and Machine Learning in Finance调试数据库编程

FINE 695

Applications of AI and Machine Learning in Finance

Homework

Overview:

The purpose of this assignment is to utilize the Machine Learning (ML) tools covered in class to analyze large datasets of stock characteristics, train models to predict future stock returns, and use this predictive information to manage a portfolio investment strategy. The effectiveness of the portfolio trading strategy should be assessed through a back-testing exercise using an Out-of-Sample (OOS) approach. Evaluation should be based on traditional performance metrics: Alpha, Sharpe ratio, average returns, volatility, drawdown, and maximum one-month loss.

Data:

You will all receive the same monthly data (homework_sample_big.csv) for large-cap US publicly traded companies from 2005 to 2023. Each stock each month has 145 characteristics, which serve as your right - hand side predictors (factor_char_list.csv). Their definitions and acronyms are described in the table in Appendix A at the end of this document. Note that all predictors are lagged by one month from time t, while all returns, the left-hand side variable (stock_exret), are from time t+1. Thus, this is a truly predictive exercise. The data are on my courses, HW folder.

Your approach to training the machine learning algorithms should follow the methodology used in class examples. Specifically, use the first 10 years as the initial training sample, the next 2 years as the validation sample, and then the subsequent year as the out-of-sample test period. You can theneither expand the window by one year or roll the training and validation samples forward by one year.

Note that the provided data includes tradable investment strategies designed to deliver  performance competitive with the market portfolio. The data guarantees the presence of signals and investment strategies based on those signals. However, the magnitudes and economic values that can be achieved within this sample are unknown. Your objective is to develop the best possible investment strategy, with no restrictions on turnover, leverage, or trading costs.

Choice of ML algorithms:

We have covered two main approaches. The first approach focuses on directly predicting stock returns. In this context, we explored penalized linear models, regression trees, and feedforward neural networks.

The second approach is based on factor investing. This method identifies latent factors in a data-driven manner and then predicts future expected returns based on these factors. Specifically, we discussed Instrumented Principal Component Analysis (IPCA) and Autoencoders.

Both approaches have their advantages and disadvantages. You are free to choose either method and are not limited to the techniques we have covered. There are many other methods available, and you are encouraged to innovate by applying alternative approaches.

Performance Evaluations:

You are developing an active management trading strategy with the objective of outperforming  the benchmark. Since you will be trading large-cap US stocks, your benchmark will be the S&P 500.

Additionally, you will compare your strategy against a market timing benchmark. You are provided with weighted bagging code for time series predictability of the S&P 500. Each month, you can predict the S&P  500 return for the following month. If you expect a high or positive return, you can invest up to 150% of your portfolio in SPY, using 50% leverage. Conversely, if you predict a negative return, you can pull out entirely and hold cash. In this scenario, you trade only one ETF (SPY) and rely on your ability to predict this time series using macroeconomic variables. You may expand the set of macro-variables.

Your task is to estimate the out-of-sample Sharpe ratio of this strategy and compare it to a passive, buy-and- hold SPY strategy, as well as your main objective: the active management strategy.

Reports/Decks

Designing an investment strategy is the holy grail of the quantitative asset management industry. Most initial attempts either fail or prove to be non-tradable in real life due to market frictions. However, your exercise is simplified as you are not restricted by market frictions, allowing your strategies to potentially deliver very high results.

We do not expect you to create the next groundbreaking algorithms like those of D. E. Shaw or Renaissance Technologies. Instead, we want you to give it your best effort and present your most promising idea. Your creativity in setting up the training and the methodology you use will be important criteria for evaluation, alongside the out-of-sample performance of your investment strategy.

Your presentation should include at a minimum the following:

1. Methodology description

2. Out-of-sample R2

3.  Table with out-of-sample performance statistics: Alpha, Sharpe ratio (annualized), average return, standard deviation, maximum drawdown, and maximum one-month loss for your main strategy compared to SPY and market timing

4. Graph of cumulative out-of-sample returns for all three strategies

Do not exceed five slides or a three-page Word document for your report. Please submit your code along with the presentation.