Update on my long-only Donchian breakout model for active stocks, mostly tech.
I built it to complement my rotational momentum strategies because it moves to risk-off much faster.
The logic is simple:
- Enter long on a breakout above recent highs.
- Trail the position with a stop.
It is currently in drawdown, but nothing dramatic so far.
Current longs: $SPCX, $AAPL, $TSLA.
I don’t have a single “best” strategy.
They all have good and bad periods. What I find useful is running them together in a portfolio.
A recent example: Fast long mean reversion on daily bars has struggled lately. Meanwhile, the slower weekly long mean reversion model I track on S&P 500 stocks recently made new equity highs.
That makes a daily/weekly combination worth testing. Even within the same strategy family, different timeframes can help diversify a portfolio.
The weekly model’s principles are simple:
- Use RSI on weekly charts.
- Favor lower market volatility.
- Trade in the direction of the main trend.
I track the model’s performance and statistics here:
crackingmarkets.com/buy-the-…
The previous maximum drawdown in this Nasdaq 100 long mean reversion backtest was about 20%. This November, it reached about 40%.
Stressful periods like this are often the best learning opportunities.
This is the model I share with a free backtester on Cracking Markets, under Interactive Models → Buying Short-Term Dips in Stocks (I dont trade this live).
The current selloff in AI-related Nasdaq stocks has pushed it into a deep drawdown. What are the lessons?
Mean reversion takes positions against a move. Sometimes that move keeps going, and it hurts. I have to plan for future drawdowns worse than anything in the backtest.
What I do in my own portfolio:
- I trade mean reversion across broader stock universes to reduce concentration. That helps, but only so much.
- I limit the combined allocation to ALL my long mean reversion strategies. Several systems buying the same selloff can suffer together.
For example, a 20% portfolio allocation suffering a 30% drawdown would mean roughly a 6% portfolio loss, before gains or losses elsewhere. That is already a serious hit. My own allocation is more conservative.
We have to accept risk to earn returns. Being smart does not remove it, and the backtest may not reveal its full extent. That is why I don’t treat the worst historical drawdown as the basis for how much risk I can afford.
In this model, the historical maximum was about 20% - until it became about 40%.
A quieter year so far for my intraday breakouts, but the equity curve is holding up quite nicely.
This is a QQQ backtest of my intraday volatility breakout, using the exact settings I share on my blog. The last few years are out of sample.
The idea is simple:
- Breakout entries at today’s open ± ATR.
- A small stop, otherwise exit at the end of the day.
- A filter to trade only days with higher potential for a larger move.
Great to see everyone at our annual Prague meetup on Saturday. 90 fellow Czech systematic traders in one room.
The biggest progress I noticed: how much faster people are developing portfolios of different strategies, using AI as a technical assistant. For traders who don’t program, the technical barrier is finally coming down.
It also got me thinking about a meetup in Málaga, Spain.
I’d be happy to organise something for a smaller group if there’s enough interest. If you’d be interested in joining, let me know.
6
27
2,296
I’m glad I scale into my Nasdaq long mean reversion trades in several legs.
This is a type of the MRZ scale-in strategy I shared here earlier. I still have room to add at lower prices and bring down my average entry price.
But yesterday’s dip was a big one.
6
14
1,956
I stopped trading this intraday long/short mean reversion strategy a few years ago. The average profit per trade had become too small for me.
But I keep coming back to it. The backtest still suggests there is an edge.
The logic is simple:
• Screen Russell 3000 stocks for large, volatile moves.
• Use limit orders the next day to trade a reversal.
• Take up to 10 long and 10 short positions.
• Exit if a wide intraday stop is hit, or at the end of the day.
Fees are included in the backtest. No overnight positions.
One thing I really appreciate about coding with LLMs: they don’t get fed up with me.
Trading research involves a lot of pivots. You put a huge amount of work into preparing data and running initial tests. Usually the results show that the idea leads nowhere. So you start again. I can understand why a programmer would eventually get tired of that.
I’m currently testing systematic options strategies. I hit a wall and needed to move the app to a new server with more storage; finish a tool for manual analysis; redo calculations we’d spent the last few days working on.
At that point, a human programmer might have told me to fuck off. The LLM had everything prepared overnight, including an independent audit. What a relief.
8
25
2,443
I still find it crazy how much a monthly rotation into the top 5 Nasdaq 100 stocks by momentum has outperformed the index over the past 10 years.
Green: momentum rotation.
Gray: Nasdaq 100.
But the volatility is wild too.
That’s why, in my own trading, I try to get the best of both worlds:
• I trade momentum rotation instead of simply holding the Nasdaq 100.
• I scale the exposure using volatility targeting, aiming for around 20% annualized volatility.
So far, I’m getting similar performance to the Nasdaq 100 with much less capital allocated.
I’m getting ready for the annual in-person “traders’ chat” I run with Czech traders, where I share what I do and how they can do the same.
Part of the preparation is refreshing the stats from last year’s presentation. The portfolio loosely follows what I trade in my own account.
The equity curve combines 2 rotational strategies, 4 long mean reversion strategies and 1 intraday breakout strategy.
Blue: out-of-sample performance since last year’s meeting.
Dark line: buy-and-hold SPY.
What a nice year.
Return +24.5%, max drawdown -4.73%, Sharpe 2.33.
And average overnight capital use of only 27%, with a maximum of 65%.
There really don’t have to be that many components to run a stable systematic portfolio.
7
23
2,313










