Demo mode — synthetic historical data

Backfilled 90-day predictions for all 13-market seed tickers using a momentum-based synthetic signal. Not the live model's output — this view shows what the analytics WILL look like once the real predictions_log has accumulated. Real-data view at the link.

Quant tool

PCA / factor decomposition

Run principal component analysis on the daily-return matrix of your watchlist. Top eigenvalues = orthogonal factors that explain joint variance. High PC1 share ⇒ your 'diversified' watchlist is really a single factor (market beta) with noise.

💡 Plain English: You think your 30-stock watchlist is diversified — but PCA can show it's really 1 hidden factor (market beta) with noise. Eigenvalue analysis reveals the hidden concentration.

Tickers

One market at a time — PCA aligns the daily-return matrix on common dates. Mixing US + TW + AU collapses the overlap window.

Vol-normalize (recommended for mixed-vol watchlists)

Methodology. Power iteration with deflation on the NxN covariance matrix. Top-K eigenvalues sorted desc; explained variance = eigenvalue / trace(cov). Vol-normalize divides each return series by its stdev so PCA reflects correlation rather than vol concentration.

PCA / factor decomposition

💡 Plain English: You think your 30-stock watchlist is diversified — but PCA can show it's really 1 hidden factor (market beta) with noise. Eigenvalue analysis reveals the hidden concentration.