VectorPeak founder proposes hierarchy adjustment to Kelly sizing

Vallikat Peethamber, founder of VectorPeak Technologies, has argued that the Kelly criterion used by many quantitative investors omits a structural variable and that a two-level market hierarchy points to a position size of 1/18, or about 5.6%. The claim, published as an external opinion on Finextra, would place common full-Kelly sizing above what Peethamber says is appropriate for mean-reversion strategies operating across nested market timescales.

The Kelly criterion, developed by John Kelly at Bell Labs in 1956, is used to estimate what share of capital to risk when a bettor or trader has a measurable edge. In its standard form, Peethamber said, the formula assumes decisions are independent, identical and flat, conditions he argues do not match markets where intraday moves sit inside daily patterns and broader regimes.

How the proposed fraction is derived

Peethamber's argument is that a binary trading decision, such as buy versus sell or long versus short, should be treated differently when it is embedded within a two-level market structure. He said the smallest integer structure that can represent both elements is 18, expressed as 2 multiplied by 3 squared. In his framework, the 2 represents the binary choice and the 3 squared represents the two layers of hierarchy.

Because those components are mathematically independent, Peethamber said the structure cannot be reduced further without losing information. He therefore identifies 1/18 as the minimum step size consistent with that structure, and treats it as the Kelly fraction for the two-level case.

On that basis, he said a strategy using the fraction would leave a residual deviation of (17/18) to the 18th power after 18 trades, or about 36.8% of the starting value. Peethamber described that as one e-folding, a standard mathematical measure of convergence, and said it implies an 18-trade cycle for mean-reversion strategies in a two-level hierarchy.

Implications for pairs trading

Peethamber linked the calculation to statistical arbitrage and pairs trading, where traders often look for instruments whose spreads revert over time. He said pairs that share the same market hierarchy should converge along a (17/18)^N curve, while pairs that appear cointegrated through conventional statistical tests may fail under stress if they belong to different hierarchy levels.

He also extended the same reasoning to an 18-day cycle for intraday strategies, saying it could explain why equity mean-reversion systems are often observed over windows of roughly three to four weeks. That assertion was presented as a structural claim, rather than an empirical backtest.

Peethamber compared his proposed fraction with common Kelly variants. For a strategy with a 10% edge per trade, he said full Kelly would set the stake at 10%, while his two-level framework would set it near 5.6%. He said half-Kelly at 5% sits close to the proposed level, though for different reasons. He also said the fraction would shift to 1/54 for a three-level hierarchy and to 1/2 for a single-level binary case.

Risk management claim

The post also set out a related risk-management constant of 1/12. Peethamber said that figure represents the maximum portion of a portfolio that can be rebalanced in one period without disrupting its internal risk structure. He attributed correlated failures in 2008 to a systemic breach of that ceiling across markets.

Finextra labelled the material as external author content and said it was published without editing and reflected the author's views. The claims have not been presented as investment advice, and traders or risk managers assessing them would need to test the framework against their own data and controls.

This story draws on original reporting from Finextra Research.