Founder
Koray Gocmen
Koray comes from a software engineering background and now spends most of his time researching, testing and running systematic strategies.
Everything published here comes out of work running with the firm's own capital. The code and the data recipe are included so the results can be reproduced rather than taken on faith.
Research by Koray Gocmen
Buying the next bitcoin cycle through MSTX calls
Bitcoin's four-year cycles have each climbed less than the last, and the decay is regular enough in log terms to extrapolate: the next run projects about 3.9x. Here is a $50,000 position built on that view, what it pays if the path holds, and the several ways it does not work. Holding a 2x daily fund costs around 73% a year, so the whole trade turns on clearing that hurdle before the calendar runs out.
The half of Borsa İstanbul that is missing from your backtest
927 companies have traded on Borsa İstanbul since 2000 and 612 are listed today. Build a universe from today's tickers and 315 firms, every one of them a company that failed, quietly vanish. Here is how large that error is, window by window, and the public archive that removes it.
Stepping aside when the trend gets ahead of itself
A trend-following book on a 3× leveraged ETF whose whole edge is the quarters it spends in cash. It turned an 82% buy-and-hold drawdown into a 34% one, which matters more than the return figure next to it.
Running two Turkish momentum books at once
Two strategies with nearly the same return and nearly the same drawdown. Run together they gave up nothing in return and took three points off volatility. But only one point off the worst drawdown, because half-correlated books fall together on the day it counts.
Ranking Borsa İstanbul on momentum, near the highs
Rank the market on trailing return and you fill the book with wrecks bouncing off the floor. Screening for proximity to the recent high first changes what the ranking selects, and it is the one part of this design that came out of research rather than convention.
Buying one-year highs on Borsa İstanbul
A breakout book that is wrong 60% of the time and holds a concentrated basket of small Turkish stocks that just made new highs. Ten positions out of 687 produced 52% of the gross gains, which is the strategy rather than a flaw in it.
Proving a backtest is not cheating
Look-ahead bias has no visual signature, and the version that matters inflates a result by less than two percentage points. There is a five-line test that catches it anyway, and a way to check the test itself works.
One sweep, six ways to pick a config
Ninety-six configurations, six explicit rules for picking one, and a holdout none of them were chosen on. The training table ranked them almost exactly backwards, the plateau heuristic finished last, and the best result came from a parameter I removed to prove it was needed.