Maynard Metrics: Base Model V1

Overview

This directory contains the initial research and modeling for the Maynard Metrics pricing algorithm. We focus on correcting the "Conservative Bias" in expert estimates and quantifying upside risk using Quantile Regression.

Data Source

The analysis relies on proprietary Sotheby's auction data (2020-2025).

• Auctions Table: Metadata for events (Date, Location, Total Sales).
• Lots Table: Item-level details (Artist, Estimates, Hammer Price).

Note: Raw database files (*.db) are excluded from this repository via .gitignore for data security and size constraints.

Notebook Structure

1. 01_data_inspection.ipynb: Auto-discovery of database schemas and data quality checks.
2. 02_quantile_regression_lgbm.ipynb: Training the LightGBM model to predict the 10th, 50th, and 90th percentiles.
3. 03_simulation_evaluation.ipynb: Simulation proving the mathematical validity of the "Conservative Bias" correction.
4. 04_summary_dashboard.ipynb: Visualization suite for generating the "Findings" charts used on the website.

Setup & Requirements

Ensure your python environment has the following installed:

pip install pandas numpy matplotlib seaborn scikit-learn lightgbm jupyter