💡
Switch to All time?
For more accurate suggestions we'll switch to "All time", which gives a better overview of your full trade history.

Backtest: version 1 vs version 2

In July 2026 the entire buy and sell logic was replaced. Both versions were run over the same year of Bitvavo data, on the same coins, over the same period. These are the results, and the raw trades are downloadable so you can check them yourself.

fees included 15 Sep 2025 - 7 Aug 2026 only coins in the daily volume top 80

This page is for you to explore, not a recommendation. See our disclaimer.

€100 per position
Version 1.0
−59%
−€586 · 1,520 trades taken, 1,076 signals skipped
Version 2.0
+21%
+€211 · 304 trades taken, 1,002 signals skipped
Bitcoin, same period
−42%
buy and hold, fees included

Move the sliders and both versions are recalculated on the real trades. A position takes a slot until it closes, so with fewer slots the bot has to skip more signals. Nothing is extrapolated here: every trade that appears was actually produced by the strategy on Bitvavo data, it is only the question of whether there was room for it that changes.

Why a bot can win 86% of its trades and still lose 59%

This is the part worth understanding, and it is what the rebuild was about. Both versions win the large majority of their trades. What separates them is the small group that goes badly.

Result of the trade V1 trades V1 total V2 trades V2 total
more than 20% down 97 -2917 18 -594
10 to 20% down 5 -78 8 -101
0 to 10% down 109 -64 6 -1
0 to 1% up 504 +301 49 +33
1 to 2% up 422 +597 66 +98
2 to 5% up 305 +911 103 +335
more than 5% up 78 +664 54 +440
Version 1 made 1,520 trades and version 2 made 304. In both, roughly 6% of trades ended more than 20% down, and in both those trades cost about €30 each. The difference is volume: five times as many trades means five times as many of those. Version 2 trades less, holds a little longer, and its winners are bigger. Nothing here is a clever trick, it is a strategy that says no more often.

Check it yourself

Every trade of both versions, with entry, exit, holding time and result. Same columns in both files.

⇩ Version 1 1,520 trades · CSV ⇩ Version 2 304 trades · CSV

What this is not

A backtest is not a guarantee for the future. This is one year, in a market where bitcoin fell 42%, and a strategy that does well in a falling market can do poorly in a rising one. Version 2 needs 35 days of history before it can compute anything, which is why the test starts in September and not August. And the results assume every order fills at the closing price of its candle; in reality there is a delay of up to a minute between signal and order.

The bot currently runs in paper mode for everyone, which means no real money, precisely because one good year on paper is not the same as a good year live.