How we test, and what the latest tests found

Every figure in this note assumes no trading costs unless a paragraph says otherwise. Fees, foreign-currency charges and tax would lower them.

Last time I told you my own strategy did not reliably beat buy-and-hold. Since then I have run harder tests. Some results are good news. Some are not. This post gives both, because a test you only report when it passes is not a test.

Why the first way of counting wins was too generous

The early numbers counted a win whenever a strategy setting beat buy-and-hold on a stock. By that count wins looked plentiful: 4,091 of 9,198 tests won. But only 957 of them were on stocks where buying and holding actually made money. Many wins were simply the strategy sitting in cash while the stock fell.

The stricter rule, and how rare it is

So I wrote a stricter rule down before looking at any names. The strategy must beat holding over the whole history, in at least 70% of rolling five-year windows, and in at least six of the ten years from 2016 to 2025. One fixed setting, nothing tuned. On 1,022 stocks, 45 pass with the pullback strategy. That is 4.4%. As a check I ran the same rule with the entry timing scrambled, and 3 stocks passed that, so a few of the 45 will be luck.

That number is not a chance of making money. It is the share of stocks that pass a hard test.

What the newest tests found

Time out of the market: a strategy that holds cash half the time can beat holding in a falling market without any skill. I tested for this by comparing each strategy with a plain portfolio that holds the same average share of the stock every day. On 400 stocks I had never touched, no strategy passed more often than luck alone would explain. The most that passed was 14 stocks for one strategy, against 20 expected by luck at the bar I set in advance. My conclusion: beating holding is easy to do by sitting in cash, and it does not show good timing.

Stocks that no longer exist: my price data mostly drops companies that were delisted. That flatters every strategy. I tested the pullback strategy on 192 delisted stocks, using the earlier cost assumptions, and none of them passed the trader test. I have not yet repeated that with no trading costs, and the comparison is not like for like, so I do not read it as proof of failure either. It does mean the survivors cannot be taken at face value.

The five headline stocks

Five stocks are still at the front: PRU.L, BANFP, BTO.TO, APA and ALLAN.PA. I found them by looking at their own history, so they are a starting point, not a result. On prices to 2 October 2026, the pullback strategy made a median of 10.7% a year on them, against 6.7% for holding. All five pass the rule.

From now on the only fair test is what they do next. A forward test has just started on them. It needs 126 trading days before I will read it, and I will report it then, whichever way it goes.

What I can and cannot say

I can say that a fixed, untuned rule beat holding on these five in most windows and most years in the past. I cannot say it will keep doing so. I cannot say it is statistically significant once I count how many things I tried. I cannot say it works on stocks that were delisted. This is research, not financial advice.

What I will do next

I will keep running new tests and keep the old results next to them, so you can see what changed and why. If the five stop holding up, I will say so here.

Does the pullback strategy beat buying and holding?

Not on most tickers we test. On SPY, 3 of 553 backtests beat holding (0.5%), and the median strategy compounded at 1.3% a year against 10.8% for holding. On NVDA, 3 of 553 beat holding (0.5%), at 8.3% a year against 36.4%. These are hypothetical backtests with no trading costs, before tax.

Why publish failed tests?

A test reported only when it passes tells you little. Each card shows how many backtests were run and how many beat holding, so the failures stay next to the wins.

Do the figures include trading costs?

No. Every card and backtest assumes zero trading costs. Fees, spreads, currency charges and tax would lower the results.

Is this financial advice?

No. This is research, not financial advice. Past results, real or simulated, do not tell you what will happen next. Capital at risk.