Using Randomness Tests to Study Market Regimes

Summary

The document considers statistical tools for testing whether historical changes in a market variable appear random, with the aim of identifying unusual patterns and consistency over time. It points to the NIST test suite, a collection originally used to assess the quality of pseudorandom number generators, as a possible source of tests for financial data.

The answer suggests an exploratory interpretation: periods with high measured randomness may correspond to stable conditions, while lower randomness could coincide with state transitions. It notes that multiple implementations exist and refers to a discussion of applying the suite to financial data. No specific test statistics, thresholds, empirical results, or validation procedure are given. Randomness tests alone do not establish that a pattern is tradable or that a detected transition has economic meaning; conclusions would depend on the selected tests, data representation, and treatment of changing distributions.

Key ideas

• The NIST test suite offers statistical tests developed to assess pseudorandom number generators.
• The suite can be explored as a way to characterize randomness in financial time series.
• The proposed interpretation links high randomness with stable periods and low randomness with possible state transitions.
• The document provides no thresholds or empirical evidence that randomness scores predict market behavior.
• Test results require context because changes in measured randomness do not by themselves establish a tradable signal.

Tags

# Tools to measure the randomness of database


# Tools to measure the randomness of database












I was working on historical data looking for anomalous patterns that we would not expect to occur at random. I'd like to create a scheme to analyze data and markets to test for statistical significance and consistency over time.

Question : Are there some tools we could use to evaluate the nonrandomness of a database? What do we use usually to say that something is completely random or not (statistically)? Figure we work with change in some value over time.

## Answer by HerbN (score 1, accepted)

https://quant.stackexchange.com/a/30396

Computer programmers use the NIST test Suite to evaluate the quality of pseudo-random number generators. I've been interested in using it to test market data for periods of stability (represented by high randomness) and periods of state transition (represented by low randomness). Multiple implementations exist.

A discussion of using the NIST suite to test randomness of financial data can be found in this article along with the author's details on his own implementation of the test suite.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.