New Highs
Trading research · tested, not told

What traders believe, put to the test

Do prices turn at trend lines? Are stops hunted? Does the trend still pay? Twelve papers take ideas every trader has heard and test them on decades of futures and stock data, with costs, out-of-sample periods and t-values. Some survive. Most do not. All of them teach something about how markets move.

12papers
38futures markets
17,065US stocks, delisted included
51years of daily data; 1-minute bars since 2008

The ledger

Twelve beliefs, twelve verdicts. The short version of each paper; click a row to read the long one.

The belief
What the data say
Verdict
“Prices turn at trend-channel lines.”
Real lines bounce no more often than the same lines shifted half a range away.
no edge
“A clean regression channel holds the next bar.”
Clean channels contain the next bar less often than ordinary ones: 80% against 88%.
backwards
“Stops beyond swing points get hunted.”
After the obvious level breaks, price runs further and returns less often: a stop cascade, not a hunt.
real, tiny
“Round numbers act as support and resistance.”
Price stops at round numbers, but does not turn because of them.
real, below a tick
“The last half hour follows the day.”
Gone in equity indices since 2020; still there in currencies, where it was weakest.
faded
“Trade the break out of the noise area.”
Sharpe ratio 1.68 before publication, 0.10 after it, and −0.99 in 2026 in ES.
decayed
“Darvas boxes pick the winners.”
A box breakout is no better than any other new 52-week high; its 1990s success was momentum.
no alpha
“Trade the opening range breakout.”
A real effect before costs; net of one tick per side none of 611 versions makes money.
real, too small
“Trend following works even with random entries.”
It replicates: 14.6% a year since 1975. But no coins are needed, and part of the profit is drift.
holds
“Pick trend or mean reversion per market.”
Asset classes differed in hindsight; a market’s own history did not tell them apart. Trend signals gain from trading at the same close.
by asset class
“Gaps get filled.”
Three in four small gaps fill, but the level on the other side of the open is reached almost as often.
chance
“Tilt a trend program toward the sectors that trend.”
Sector weights set with the data available at the time did not help, and even perfect hindsight adds little.
no gain

The papers

Each paper states its question, tests it against a null that the same rules produce on random paths, and reports the result whatever it turns out to be.

How the testing is done

The same habits run through every paper. They are what makes a negative result worth reading.

A fair null

Each pattern is compared with the same rules applied to random price paths with the same volatility, or with the same lines shifted on the real path. An effect has to beat chance, not zero.

Out of sample

Published strategies are tested on data after their publication, rules are fixed before the test, and follow-up tests are labelled as such. Where a strategy decayed, the papers say when.

Costs and t-values

Results are reported in ticks and ATRs next to realistic costs, with Newey–West, bootstrap or Ledoit–Wolf tests. An edge smaller than the cost of the trade is not an edge.

How to read the numbers: a guide to the statistics in these papers →

“The useful information is at the level of asset classes and horizons, built from many markets over many decades.”— The Right Strategy for the Right Market?