oregon-timber-market-timing

Oregon Timber Market Timing Model

Predicts Douglas fir stumpage prices ($/MBF) and recommends optimal timing for marketing timber on a 20-acre tract in Oregon, Pacific Northwest.

How It Works

1. Ridge regression with 21 engineered features (price lags, lumber PPI, housing starts, interest rates, CAD/USD, seasonality)
2. Walk-forward validation on 2015–2024 hold-out data
3. 500-sample bootstrap forecast for 8 quarters ahead with prediction intervals
4. Monte Carlo optimal stopping (50,000 price path simulations) to decide: sell now or wait

Performance

Key Price Drivers

| Rank | Feature | |Coefficient| | |------|---------|-----------------| | 1 | price_lag1q | 45.8 | | 2 | lumber_futures | 26.0 | | 3 | lumber_ppi | 25.0 | | 4 | Q1 seasonality | 20.4 | | 5 | price_lag2q | 20.1 |

Usage

import pickle
from sklearn.pipeline import Pipeline

with open('oregon_timber_model.pkl', 'rb') as f:
    bundle = pickle.load(f)
    model = bundle['model']
    feature_cols = bundle['feature_cols']

# Prepare features matching bundle['feature_cols'] and predict
# price = model.predict(features)[0]

Full training and inference pipeline available in timber_final.py.

Data Sources (Production)

Note: Training data in this repo is synthetic, calibrated to published USDA PNW stumpage statistics. Replace with actual ODF data for production use.

Citation

Inspired by Faustmann rotation model (1849), Clarke & Reed real-options timber harvesting (1989), and USDA Pacific Northwest stumpage research (PNW-GTR-423, PNW-RP-436).

<!-- ml-intern-provenance -->

Generated by ML Intern

This model repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

• Try ML Intern: https://smolagents-ml-intern.hf.space
• Source code: https://github.com/huggingface/ml-intern