Stock price predictor

What this project does

This educational project demonstrates a simple approach to short-horizon stock-price forecasting using synthetic adjusted-price time series. It emphasises chronological evaluation (walk-forward), compares a lag-feature regression model to two naive baselines (last-price and simple-return extrapolation), and reports mean absolute error (MAE) and R2 at multiple horizons.

This is not investment advice, trading execution code, or a backtest using licensed market data. Instead the dataset is a small synthetic stand-in that mimics key statistical features of equity returns for educational purposes only.

Data

The data are synthetic adjusted prices generated inside the code with a geometric Brownian motion (GBM) process. From that price series we build lag features (last price, prior returns), short moving averages and target prices at horizons 1, 5 and 10 days. For reproducibility the generator uses a fixed random seed passed to the run() function.

How to build it

1. Create the synthetic adjusted-price time series (GBM) and build a Pandas DataFrame of features and targets.
2. Prepare lag features and rolling statistics and align them so every row only uses information available at the timestamp of prediction.
3. Implement a walk-forward (expanding-window) evaluation: fit a Ridge regression on past data and predict the next evaluation block; compute MAE and R2 per horizon and compare with baselines.
4. Package results into a small JSON-like dictionary and print them from an if name == “main” block.

How to run

Run the project file with Python 3.11+. It prints a JSON object with the metrics. Example:

python project.py

You can also import run from project and call run(seed=42) programmatically.

How to test

This repository includes pytest tests. From the project root run:

pytest -q

The tests check presence of expected keys, conservative metric floors, valid ranges and determinism of the main metric when called with the same seed.

Results

Measured by running python project.py (seed 42) on 2026-10-23; the tests in test_project.py assert the floor.

Metric	Value
mae_horizon_1	1.0004
r2_horizon_1	0.9821
baseline_last_r2_h1	0.9875
baseline_last_mae_h1	1.2272
baseline_return_r2_h1	0.9886
baseline_return_mae_h1	1.1786
mae_horizon_5	2.3329
r2_horizon_5	0.8940
baseline_last_r2_h5	0.9636
baseline_last_mae_h5	2.0488
baseline_return_r2_h5	0.9411
baseline_return_mae_h5	2.6039
mae_horizon_10	3.2970
r2_horizon_10	0.8097
baseline_last_r2_h10	0.9389
baseline_last_mae_h10	2.7587
baseline_return_r2_h10	0.8342
baseline_return_mae_h10	4.5031
avg_transaction_cost_per_prediction	0.0500
note	Synthetic GBM used. Model: Ridge regression. Chronological walk-forward evaluation. Educational only.

Limitations

• The dataset is synthetic and small. Real market data include corporate actions, market microstructure effects and regime changes that are not modelled here.
• No transaction or slippage modelling beyond a simple illustrative cost parameter; a proper backtest requires order execution modelling and realistic costs.
• The ML model used here is a simple Ridge regression; more advanced models and careful feature engineering may be required in real applications.
• Chronological validation here is simplified; production systems should use more thorough walk-forward hyperparameter tuning and out-of-sample monitoring.

Next steps

• Replace the synthetic series with licensed historical adjusted prices and ensure timestamps for feature availability are recorded.
• Extend features (fundamentals, intraday liquidity measures) and add robust cross-validation for time-series (nested walk-forward).
• Implement a realistic trading-cost model and a portfolio-level evaluation (Sharpe, drawdown) if the goal is strategy assessment.
• Experiment with more expressive models (gradient boosting, recurrent models) while guarding against overfitting and data leakage.

---

Full case study, build steps and evaluation checks: Stock price predictor on networkershome.com

Built by Networkers Home · Learn the skills in the AI Engineer Course · Part of 100 AI projects with code