barj capital

1 · Published

… strategies, each published as one that works

The … strategies of Kakushadze and Serur's 151 Trading Strategies and … more rebuilt from the academic literature. Each point is one of them.

What is left of them once they meet real data and a real broker?

2 · Point-in-time data

Rebuilt with only what was known at the time

CRSP, Compustat, the Jensen-Kelly-Pedersen factors, futures, options and FX, with every universe chosen from what was known on the date: no survivor picked with hindsight. … cannot be rebuilt from the data we hold. Of the … that can, … show a gross effect before any cost.

$$t_{\text{NW}}=\mathrm{SR}\sqrt{T}\ge 2$$

Across: gross Sharpe ratio. Up: turnover per year, log scale. SR annualised, T in years; the t-statistic is Newey-West corrected.

3 · Interactive Brokers costs

Charged the broker's own fee grid

At …, the largest account we test: commissions, exchange and clearing fees, spreads, square-root market impact, borrow on shorts. Every point slides left by what it pays, and the more it trades the further it slides: the median gross effect gives up … of Sharpe.

$${\mathrm{SR}}_{\text{net}}=\frac{\mu -\tau c-b}{\sigma }$$

… real effects are killed by costs. … survive, tested together as one family of …: a t of 2 kept by Benjamini-Hochberg at 5%, or robust, a t of 3 with a Romano-Wolf p below 0.05.

τ turnover per year, c all-in cost of one unit traded at that size, b borrow.

5 · What is left

… survive at …, … of them robust

… of the … published.

Net Sharpe at … over the whole history. Gold: robust; green: significant.