Research index
Free editorial articles on how AI models forecast volatility and represent uncertainty — written from Taipei, read anywhere.
The archive collects our editorial articles on AI in investment. Each entry explains one method for forecasting volatility or for representing uncertainty, in plain language with the assumptions stated out loud. Everything below is free to read; nothing is gated, priced, or tied to an account. New entries are added as we finish them.
Article 01 · Volatility
A walk through the workhorse volatility model — its persistence parameter, its mean-reversion to a long-run variance, and the regime shifts that make its forecasts lag reality. We show the calibration checks that reveal the lag before it costs you trust.
Article 02 · Uncertainty
Why the options market steepens its skew ahead of scheduled announcements, what the skew actually prices, and how to read it as an uncertainty signal rather than a directional one. Includes the difference between implied and realised volatility as a sanity check.
Article 03 · AI methods
How machine-learning hybrids combine GARCH features with order-flow and macro inputs, where they improve on classical forecasts, and the training-window problem that makes them overconfident in calm markets and silent in stressed ones.
Article 04 · Uncertainty
A practical introduction to conformal prediction as a way to produce prediction intervals with a coverage guarantee — and an honest discussion of what that guarantee assumes and when it stops holding under distribution shift.
Article 05 · Volatility
How realised volatility, bipower variation, and jump-robust estimators turn high-frequency returns into a ground truth for evaluating forecasts — and the microstructure noise that distorts them at the finest sampling intervals.
Article 06 · Uncertainty
The distinction that gets blurred in market commentary: what a Bayesian credible band says about a parameter, what a frequentist confidence interval says, and why mixing the two produces uncertainty numbers that no one can defend.
Propose a topic. The clearest reader questions tend to become the next entries in this index.
