Five market hypotheses, tested to the end

Five independent studies, each with its own hypothesis: an empirical replication of Fama & French (2004) — the survey “The CAPM: Theory and Evidence”, not the 1993 three-factor model —, with the GRS test and the Betting-Against-Beta anomaly; Volume Profile put through falsification across 17 to 33 years of data; an expectancy backtester with R-multiples and risk-of-ruin computation; the GOLD11 premium arbitrage on the Brazilian exchange; and the constant-dollar rebalancing rule. All public, all with a PDF report.

The question in each one was the same: does this work? Not “how do I make this work”, which is the question that produces a beautiful backtest and a real loss. So the protocol is identical across all five: walk-forward validation out of sample, results always net of transaction costs, no access to future information at any point in the computation, and adversarial tests — permutation and bootstrap — to separate signal from the particular shape of noise in that stretch of history.

What makes the set worth having is what happens when the answer is no. A technique that doesn’t survive costs is a conclusion, and it is exactly the conclusion the experiment was built to produce with confidence. The work isn’t to manufacture a positive result; it’s to reach a verdict that stays true outside the computer. Finding out a popular rule doesn’t hold up is cheaper now, in a repository, than later, with money.

So what these studies deliver isn’t a strategy. It’s the method and the conclusion — with the data, the code and the report open, for anyone who wants to rerun the test and disagree with evidence.