Skip to content

Poster

Title: Personal portfolio predictor

Introduction

The "Personal Portfolio Predictor" uses an autoencoder neural network to predict changes in institutional investors' stock portfolios. By focusing on these key market players, who hold a significant market share, the project aims to understand and anticipate shifts that can notably influence stock movements. This approach provides a straightforward yet effective tool for deciphering market trends and investor behaviors, offering an advancement over traditional methods like random guessing. It represents a step towards more data-driven and insightful investment strategy analysis.

Methodology

Data Collection

  • SEC EDGAR Database: Historical Form 13F filings were used to analyze institutional investor stock holdings.
  • EOD Stock Market Data: End-of-Day stock market data was collected to correlate with the investor holdings and understand market trends.

Data Analysis and Profile Creation

  • Investor Profiles: Developed from Form 13F data to map out historical investment patterns of institutional investors.
  • Market Trends Analysis: EOD data was analyzed to identify key performance indicators influencing stock movements.

Model Development

  • Autoencoder Model: A basic Autoencoder neural network was constructed, specifically designed to predict buying or selling actions of institutional investors.
  • Model Architecture:
    • Input dimension tailored to include investor and stock information.
    • Output dimension focused on binary decisions: buy or sell.
    • The model utilized a masked binary cross-entropy loss function to selectively focus on significant trading actions, ignoring periods of inactivity.

Model Training

  • Training Process:
    • Training data included historical investor holdings and corresponding market data.
    • The model was trained using the function, which processes data up to a specified year and quarter.
    • The function employed a custom loss function, masked_binary_crossentropy, to focus on instances where investors made definitive buy or sell decisions, excluding periods of no action.
    • EarlyStopping was implemented to prevent overfitting, with validation loss monitoring.

Prediction and Evaluation

  • Objective: To predict whether an investor will buy or sell a particular stock.
  • Simulation: The model was used to simulate future investment decisions based on historical trends and current market conditions.

Results

Our model's performance was evaluated using several key metrics, each providing insights into different aspects of its predictive accuracy:

  • Mean Absolute Error (MAE): 0.4985

  • Mean Squared Error (MSE): 0.4985

  • F1 Score: 0.0950

Interpretation

  • These metrics indicate that while the model is capable of making predictions, there is significant room for improvement, especially in its ability to accurately classify buy/sell decisions. The relatively high error metrics (MAE and MSE) and low F1 Score point towards challenges in capturing the complex patterns inherent in stock market behavior and investor decision-making.

Future Work

  • Future iterations of the project will focus on model refinement, possibly exploring more complex neural network architectures or additional feature engineering to enhance prediction accuracy. Further tuning of the model parameters and training process could also address the current limitations.