Random Walk Benchmark

The random walk treats successive price changes as independent and unpredictable — the position implied by the Efficient Market Hypothesis. As a benchmark it is the minimum bar a Markov Chain Trading Model must clear: for an N-state classifier, random-chance accuracy is 1/N (e.g. 50% for up/down, ~16.7% for six states). Aronsson Folkesson 2023 found a single Markov chain matched this benchmark and an ensemble beat it only marginally; Do Carmo 2017 beat it on a reconstruction-error metric. It appears in this vault as the primary reference point for grading Markov-chain predictive claims.

Connections

• Markov Chain Trading Model — compares_benchmark, source: https://arxiv.org/pdf/1803.06653