Methodology & About

Stocks Like These is a quantitative research tool that identifies stocks exhibiting similar historical price patterns using multiple statistical algorithms.

1. Similarity Algorithms

Correlation

Pearson correlation on daily log returns over multiple sliding windows (21, 63, 126, 252 days).

  • Fast and cross-sectionally stable
  • Captures directional co-movement
  • raw value in [-1, +1] (higher is more similar)

Dynamic Time Warping (DTW)

Non-linear alignment of z-normalized log-price paths.

  • Scale and volatility independent
  • Finds structural chart shape similarities even with slight time lags
  • raw value is DTW distance ≥ 0 (lower is more similar)

Feature Matching

Statistical fingerprint matching across 5 Key Performance Indicators (KPIs).

  • Volatility, mean return, skewness, max drawdown, & efficiency ratio
  • Z-scored across the universe and compared via Euclidean distance
  • raw value is feature distance ≥ 0 (lower is more similar)

2. Scores vs. Raw Values

Each candidate stock returns two metrics: a relative score and a method-native raw value.

score (Relative)
Universe-relative similarity score calculated as the sigmoid of the candidate's cross-sectional z-score. It expresses how strongly a candidate stands out relative to the rest of the searched universe, enabling direct comparisons across different algorithms.
Note: Because scores depend on cross-sectional universe statistics, small universes (N < ~30) produce noisier relative scores.
raw (Absolute)
Method-native affinity metric representing absolute similarity without reference to the rest of the universe.
  • Correlation: Pearson correlation coefficient \(r \in [-1, 1]\).
  • DTW: Accumulated time-warped distance (0 = identical normalized shape).
  • Features: Euclidean distance across z-scored statistical feature vectors.

3. Quality Gate & Result Filtering

Why are some results greyed out or empty?

By default, Stocks Like These applies an absolute quality gate on raw metrics, calibrated against the full S&P 500:

  • Correlation threshold: r ≥ 0.60
  • DTW threshold: dist ≤ 0.45
  • Feature threshold: Ungated (no natural cutoff across sectors)

When searching in Auto (Gated) mode:

  • Matches that fail the quality floor are styled with a muted font and a Below threshold badge.
  • If no stocks in the universe satisfy the quality floor, the result list may be empty.

Financial Disclaimer

For research and educational purposes only. Not investment advice. Past price behaviour does not predict future returns. Market data is retrieved via yfinance.