Scribd
Upload
43 views

Beating The Market:: A Simple Model To Predict Asset Returns

This document proposes a data-driven machine learning model to predict stock returns without assumptions. It finds that a strategy of buying stocks predicted to have gains and shorting those predicted to have losses outperforms the S&P 500, turning $1 into $8 compared to $2.5. However, the model has limitations, failing recently due to increased volatility causing false predictions to rise to 50%. Next steps include enhancing the model with additional features and backtesting during past crises.

Uploaded by

Nitro Nexus
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
43 views

Beating The Market:: A Simple Model To Predict Asset Returns

This document proposes a data-driven machine learning model to predict stock returns without assumptions. It finds that a strategy of buying stocks predicted to have gains and shorting those predicted to have losses outperforms the S&P 500, turning $1 into $8 compared to $2.5. However, the model has limitations, failing recently due to increased volatility causing false predictions to rise to 50%. Next steps include enhancing the model with additional features and backtesting during past crises.

Uploaded by

Nitro Nexus
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Beating the market:

A simple model to predict asset


returns

Cory and Xin 1


Motivation

● Stock market is volatile and unpredictable


○ Current Asset Pricing models fundamentally flawed
○ Assumption driven (distributional and behavioral)
○ Some empirical prediction worse than random walks
● Goal:
Without assumptions, develop a data driven model to predict stock returns.

2
Project-specific questions
At any given time, t , what’s a simple strategy to maximize return on a portfolio in the
following period?

Question road map:


1. Build machine-learning model to predict y_{t+1} (return)
2. What goes in X_t: feature set, all observable at time t
3. Based on the model, what’s the strategy and its performance?

3
Data Sources
- Data Sources (2010-2019):
- (Daily) stock prices in S&P 500 constituent
- (Quarterly) balance sheet, income statement, earnings announcement
- (Fixed) company factsheet: industry, location

- Feature Set:
- Technical Features (generated from trailing price and volume), e.g. RSI
- Fundamental Features (generated from trailing quarter), e.g. liabilities
- Dummy Features

- Summary (aggregated monthly), >500 features, 50K observations


4
Model and Prediction
- Y: Next Month Return 3 Clusters:
- Gain (> 5%)
- Loss (< - 5%)
- Neutral

- X_t (PCA, Kernel, Polynomial, …)

- Classification Models:
- Logistic
- Random Forest
- SVM
- Ensembling

- Average testing accuracy: 76%


- Worse Case: 11%
5
Model-derived Strategy Vs. Index
Strategy:

Buy predicted gain

Short predicted loss

Adjust position every


month

Realized
Performance:

Strategy: $1 >> $8
S&P: $1>> $2.5

6
Model Caveat Trade
Fed
NK war
Rate
tension
Hike Predicted Gain, Lost Big

- Volatility in recent years


caused our model to fail.

- Propensity of false prediction


ratched up to 50%.

7
Next Steps
Limitation
● Model was trained on bull market

Potential Work:
● Add news sentiment and Macroeconomic features
● Backtest on previous 2008 crisis
● Update data and re-train model parameters

8
Questions?

You might also like