Money

Strota backtested its own Darvas box screen: no edge once you pay the fees

By Strota Newsroom · 2026-08-08 · How Strota reports

backtestdarvas boxbreakoutnifty 500screensnegative result

Across 9,760 signals over six years of Nifty 500 data, the breakout the screen publishes nightly averaged -0.29% as it would actually be traded — and the friendlier long-horizon numbers answer a different question.

A breakout screen makes an implicit promise. It says: these stocks have just done something that matters. The Darvas box breakout is one of the oldest versions of that promise on a chart — a stock coils inside a tight range for a few weeks, then closes above the top of that range on unusually heavy volume. Traders have been buying that pattern for decades. Strota publishes a list of Indian names that qualify, refreshed every night. So the fair question, and the one a reader actually asked, is whether buying the names on that list would have paid.

It would not have. Tested against the exact rule the page uses, the setup produced an average loss of 0.29% per signal after trading costs, with only 38.2% of signals finishing in profit — that is the win rate, the share of trades that made money at all. Fewer than four in ten worked, and the average of all of them was negative. The pattern's own round-trip cost is enough to erase whatever small pull it has.

Here is how that was measured. The universe was 473 of the Nifty 500 names that had usable price history, over 2020-07-31 to 2026-07-31 — six years, 1489 trading sessions. Every entry and exit was charged 0.24% for the round trip, roughly what a retail delivery trade costs in brokerage, taxes and exchange fees. That produced 9,760 separate breakout signals, which is a large enough sample that the result is not an accident of a handful of trades.

The detail that makes this worth reading is narrower than the sample size. The rule used in the test is not a reconstruction of the screen — it is the screen's own rule: a box built from the prior 15 sessions with the current day excluded, a box no wider than 12% of its own top, a close that clears the top, and relative volume of at least 1.5 times normal. What was measured is therefore what the page actually publishes, not a tidy academic cousin of it.

Now the part that has to be said carefully, because several different measurements disagree with each other. Held for five sessions, the average signal returned -0.08%, with 46% winners. Held for twenty-one sessions, +1.64% and 53% winners. Held for sixty-three sessions, +6.03% and 57% winners. Stretch the holding period and the numbers turn positive. That looks like a rescue, and it is not one.

The reason it is not a rescue is that the longer horizons answer a different question. The screen lists a name for roughly one session; buying the breakout and exiting when it drops off the list is how the signal is actually traded, and that is the negative number. A multi-week hold is no longer the breakout — it is just owning an Indian equity through a rising market. And that shows up in the comparison: +1.64% over twenty-one sessions sits against a Nifty drift of about a percent a month, and among fifteen screens tested it ranked 12th.

One comparison settles it. A screen that does nothing more than flag stocks trading above their 200-day moving average — a plain "this thing is in an uptrend" filter, with no pattern recognition in it at all — returned +1.66% over the same twenty-one sessions. Statistically that is the same number. All the box geometry, the range tightness, the volume confirmation: none of it added anything beyond the observation that the stock was already trending up.

A fourth measurement in the run looks catastrophic and should be discounted. Treating the screen as a portfolio that rebalances daily into every listed name produces a compound annual return of -34.9% and a maximum drawdown of -92%, meaning the equity curve fell that far from its own peak. But that lens is unfair to a one-day event screen: rebalancing every session into a list that turns over completely is nothing but churn. The gap-up screen, which posted the strongest forward numbers of anything in the run, comes out at -33.4% under the same treatment. The figure measures the accounting, not the setup.

Across the fifteen screens, one earned its keep. A trend-template screen built on Minervini's criteria returned a portfolio CAGR of 20.8% against 14.1% for simply buying and holding the Nifty, with a maximum drawdown of -28% and around 72 names in the book at a time — and because it describes a persistent state rather than a one-day event, the portfolio lens is a fair test of it. An oversold-RSI screen came in at 14.3% and the above-200-day screen at 9.5%, both at or under the index. On forward returns the gap-up screen led, at +4.12% over twenty-one sessions with 60% winners. Everything else in the bullish column sat inside the market's own drift.

There is also a timing gap that matters to anyone reading the list during market hours. What was tested is the end-of-day version. During the session the live page recalculates from running prices, treating the current print as the day's close, so a name can qualify at 11:00 and no longer qualify by 15:30. A signal seen and acted on intraday is not the signal this test measured.

The limits on all of it are real and worth naming. The pool is today's Nifty 500, which means companies that were delisted or blew up over the period never entered the sample — survivorship bias, and it flatters bullish screens rather than the reverse. The window from 2020 to 2026 was mostly a rising market, so any long-only screen collects drift for free. And screens that need something other than price were left out entirely, though open-interest buildup has already been shown to have no next-day edge across 84k observations, and circuit-hitting stocks were settled the same way.

The finding stands as a negative one, and it is Strota's own screen it lands on. The Darvas box breakout, as the page defines it and as it would be traded, did not pay over these six years. That is a statement about what already happened, not a forecast about what a breakout will do next — but it is the kind of thing worth knowing before treating any nightly list as a reason.

Strota backtested its own Darvas box screen: no edge once you pay the fees

More money stories