Stock Forecasting Using Automated Parameter-Optimization for Prophet Models and Time-Aware Bellman-Ford Strategy Planning

Overview

This project forecasts stock prices using a combination of:

• Globally-optimized Prophet models and parameters by minimizing a custom multi-metric objective function.

• Time-aware Bellman-Ford algorithm to optimize multi-step investment strategies dynamically across different stocks.

• XGBoost as an alternative engine for tabular modeling. (under development)

• Margin analysis and profit range estimation.

It supports automated model optimization, backtesting (train test split), plotting of forecast paths, strategy visualization, and margin viability analysis.

---

Remarkable Results

The results and statistics provided below were obtained with no human intervation. The only input to the code is the stock ticker list.

• 1-month stock forecasting: Achieved mean R² = 0.92(median: 0.95) and MAPE = 9.86%(median:7.64%) over 50+ tickers using globally optimized models.
• 6-months stock forecasting: Achieved mean R² = 0.90(median: 0.94) and MAPE = 11.57%(median:8.51%) over 50+ tickers using globally optimized models.
• 1-year stock forecasting: Achieved mean R² = 0.93(median: 0.95) and MAPE = 11.80%(median:8.33%) over 50+ tickers using globally optimized models.
• 5-year stock forecasting: Achieved mean R² = 0.92(median: 0.95) and MAPE = 11.45%(median:8.66%) over 50+ tickers using globally optimized models.

Features

• ✅ Meta-optimized time series models. Works in parallel.
• ✅ Fast data loading with local caching
• ✅ Dynamic Bellman-Ford investment strategy planner
• ✅ Visualization tools
• ✅ Margin borrowing analysis with profitability check
• ✅ Easy restart and resume of stock forecasting
• ✅ Compatible with large-scale multi-ticker analysis

---

Installation

• Clone the repository:

git clone https://gitlab.com/bcucco/stock-forecasting.git

cd stock-forecasting

• Install dependencies:

pip install -r requirements.txt

---

Quick Start

• Initialize the forecaster:

from forecaster import StockForecaster

tickers = ["AAPL", "MSFT", "NVDA", "AMZN"]
stock = StockForecaster(tickers, forecast_horizon="1y")

• Download and cache stock data:

stock.data(api="yfinance", interval="1mo")

• Train models using global optimization:

stock.model(engine="prophet", metaopt=True, cores=10, maxiter=100)

• Run forecasts:

stock.run()

• Generate best investment path using time-aware Bellman-Ford:

stock.bellmanford(forecast_horizon=["60d", "120d", "240d", "365d"], invest=10000.00)

• Plot investment path:

stock.plot_bellmanford()

• Assess margin borrowing:

stock.assess_margin()

---

Outputs

• results_forecast.csv: Forecasted prices, R², and MAPE per ticker.
• bellmanford_results.csv: Top dynamic investment strategies.
• bellmanford_margin_analysis.csv: Margin viability report.
• Plots:
  Forecast curves: forecast_plots/
  Strategy graphs: plot_graph.png
  Path-wise Bellman-Ford forecasts: bellmanford_plot_*.png

---

Project structure

.
├── forecaster.py                   # Main forecasting code
├── README.md                       # This file
├── requirements.txt                # Contains dependencies to be installed
├── data_cache/                     # (Generated) Cached historical stock data
├── forecast_plots/                 # (Generated) Forecast plots
├── results_forecast.csv            # (Generated) Forecast results per-ticker
├── bellmanford_results.csv         # (Generated) Top strategy paths
├── bellmanford_margin_analysis.csv # (Generated) Margin assessment
├── LICENSE                         # Code's License Information

---

License

This project is protected under the terms impose on its license. See the LICENSE file for more details.

Author

Code developed by B. Cucco.

---