Blog

Notes on time series, reinforcement learning, and applied machine learning.

• Driving through underground rocks

  A particle filter that cannot audit its own error, and what it took to estimate that error with simulation-based inference instead.

• Can we really get alpha from market data?

  Efficient Market Hypothesis, Micro Alphas, and why probabilistic forecasting matters for turning signals into positions.

• What works for forecasting macro economic series with deep learning?

  Data quirks of macro series, which model families work (and which don’t), and why it’s rarely one-size-fits-all.

• Could multivariate time series have their own representations?

  Identifiable innovations, diagonal dynamics, and iVDFM: factor recovery, interventions, and probabilistic forecasting.

• Can we make a more risk-aware portfolio agent from utility theory?

  Recursive (Epstein–Zin) utility with Monte Carlo certainty equivalents in PPO/A2C, on Korean ETF splits.

• Effective Bird Sound Classification

  Mel spectrograms and EfficientNet for bird sound: why the mel scale helps and how to keep the pipeline simple.

• Creating and Evaluating Synthetic Tabular Data

  Sequential synthesis for tabular data, plus three checks: propensity scores, CI overlap, and quasi-identifier risk.