I’ve been trading for ~30 years. First half: fully discretionary, living inside futures microstructure. It worked—until algos started exploiting the same patterns and reacting in microseconds. Edge decay was real. So ~10 years ago I switched to systematic. Now I run many uncorrelated strategies in parallel without babysitting screens all day. I wouldn’t go back. My biggest unlock: reusability of know‐how. When I finish a new system, I plug it into a ready workflow in minutes. It monitors itself; I move my brain to the next big thing. Here’s the playbook I wish I had from day one: - Framework (design once → reuse forever) - Data → clean, feature, label. - Hypothesis → simple, testable edges (breakouts, momentum, mean reversion). - Validation → IS/OOS, realistic costs/slippage. - Risk → position sizing, max heat, portfolio exposure caps. - Deploy → automated orders, fail‐safes. - Monitor → health dashboards, kill‐switch rules, mobile app. - Iterate → new systems slot into the same pipeline. Principles that compound: - Many small, independent edges > one “genius” setup. - Process beats prediction. - Shipping beats perfecting. Discretionary taught me markets. Systems gave me scale.

A few ticks of slippage can trigger a stop-loss that the backtest never hit. That is why I don’t judge a strategy by one attractive equity curve. The chart shows hundreds of backtest simulations of my algorithmic strategy selling 0DTE vertical spreads. Each run introduces small randomized changes to execution and trade management, including entry and exit levels. A small change can produce a very different trade outcome. Over shorter periods, those differences can add up to noticeably different equity paths. But over the full backtest the longer-term picture stays broadly consistent. The strategy remains profitable across the variations I tested. That is what I like to see: not identical results, but an edge that does not disappear when I change the execution assumptions. I want a strategy that can tolerate imperfect execution - not one that needs everything to happen exactly as backtested.

Where do I get affordable intraday data for testing stocks, ETFs and options? There are plenty of data vendors. But if you want good data cheaply, the list gets much shorter. My first choice is usually broker APIs. Two I use: • Alpaca - historical data available on the free tier • TradeStation - very useful historical data if you already have a funded account I compared TradeStation, Alpaca and Polygon minute data side by side. The differences were very small. Full comparison: crackingmarkets.com/comparin…

What a month for my intraday volatility breakout system. Finally, a new equity high for the version I share on my website. This is what trading a breakout system can look like: plenty of losing trades, stretches of going nowhere, then a few big winners that move the equity curve. That is one reason I like to automate it. The last thing I want is to sit through the losses, lose patience, and miss the trades that make the difference. I track the out-of-sample results publicly, so you can see the frustrating periods as well as the new highs: crackingmarkets.com/intraday…

Breakouts around the open can be very profitable. I tested this simple NQ model from 2018–2026. $100k → $3.77M, 52% CAGR, 1.40 Sharpe, -35% max DD QQQ over the same period: 19% CAGR, 0.84 Sharpe, -36% max DD MNQ commissions + slippage are included. And here are the full rules: 1. Entries TR1 = true range of the previous full session. Long: Open + 0.25 × TR1 Short: Open - 0.25 × TR1 Orders active 09:30 - 13:00 ET. 2. Stop 0.25 × TR1 from entry = 1R. 3. Break-even After a completed bar reaches +0.50 × TR1 = +2R, move the stop to entry from the next bar. 4. Exit Stop, break-even, or market exit at 15:55 ET. No profit target, no trailing stop. 5. Re-entry Re-entry is allowed after a stop-out once price trades back through the entry level from the other side. Maximum 3 entries per day across long + short. 6. Sizing Risk 1% of current equity per trade using MNQ. Equity compounds from $100k. 7. Filters None. Costs $0.95 round-trip commission per MNQ + 1 tick slippage per side on every fill. One important observation: The strategy has performed much better during the higher-volatility years. In quieter periods, failed breakouts and trading costs create much more drag. That is worth investigating further. But as a simple, symmetric breakout with essentially no filters, I find the result very interesting. Full rules above. Test it yourself.

I see traders hunting for very short-term “seasonal tendencies.” Find the right weekday, hour and filter. Get a nice backtest. Call it an edge. I’m skeptical. Even when it stays profitable out of sample. Take this NQ example: It shorts on Wednesdays at 14:00 CT when price is below the day’s open, then exits at the close. Simple rule. Attractive equity curve. Still profitable in the untouched 2022–2026 period. But it was selected from 3,510 combinations of weekdays, times, filters and direction. And that was only the final search, after trying different testing windows and ranking methods. “But it worked out of sample.” Yes. That matters. But it does not establish that Wednesday adds anything. Maybe weak afternoons tend to stay weak, regardless of the weekday. And out-of-sample profits can happen by chance, too. Across enough searches, some rules will look good in both periods without having a genuine edge. This is why I see little value in endlessly searching the same historical data. More combinations. More adjustments. More stories to explain the winner.

Two simulated strategies. Both average 20% a year for 20 years. One ends with twice the money. And a much smaller drawdown. The difference? Volatility. Both use the same underlying shocks, scaled differently. Both have an arithmetic average annual return of exactly 20%. Strategy A: 10% annual volatility. $1 becomes $35.70. 19.6% CAGR. Maximum drawdown: −12%. Strategy B: 30% annual volatility. $1 becomes $17.70. 15.5% CAGR. Maximum drawdown: roughly −70%. How does the same average produce such different outcomes? Because returns multiply. Lose 50% one year, then gain 50% the next. Your average annual return is 0%, but your account is down 25%. That is volatility drag. The arithmetic average is not the rate your capital compounds at. This is why I care about portfolio volatility, not just average returns. Volatility targeting is one tool I use to manage it. Reducing volatility is not just about making drawdowns easier to tolerate. If I can reduce the swings without sacrificing too much average return, I can finish with more capital, not just a more comfortable ride. Your account compounds. It does not average. Illustrative simulation, not live results.

Same logic. Same settings. Different universe. Two days ago I shared the updated OOS results for my Nasdaq long mean-reversion model. The question I got: does the same logic work on S&P 500 stocks? So I changed only the universe. Using a survivorship bias free database with historical S&P 500 constituents. Results: OOS since 1/1/2024 • CAGR: 42.12% • Max drawdown: -10% • Sharpe: 1.79 Full backtest since 1999 • CAGR: 25.32% • Max drawdown: -21.83% • Sharpe: 1.28 Average capital exposure: 25.75%. Compared with the Nasdaq version, exposure is higher, but so is CAGR, while the OOS drawdown remains around the same level. I published the complete rules + interactive backtester for free here: crackingmarkets.com/buying-s… Free registration is required so the material is not openly indexed and scraped by AI bots.

This can happen to anyone. A strong live track record does not guarantee that a trading system will keep working. The chart shows FWPD on Darwinex. I have no association with it. It performed well for more than a year, then entered a prolonged drawdown that gave back most of its gains. That does not automatically mean the system is broken. But imagine the psychological pressure of going through this after bringing investors on board. You are no longer dealing only with your own doubts. You also have other people asking whether the system still works. My solution is not to rely on any single trading logic. Any system can stop working. Any system can spend years in a drawdown. A convincing backtest and a strong live start do not remove that possibility. That is why I trade more than 15 different strategy logics in one systematic portfolio and keep developing more. Diversification does not eliminate drawdowns. But I do not want the future of my trading to depend on one idea continuing to work.

Beating Nasdaq 100 buy-and-hold with only 15% average capital exposure. And the rules are public. My long only Nasdaq mean reversion strategy has added some nice gains in September. Over the last few years, out of sample: • CAGR: 35% • Max drawdown: −11% • Sharpe: 1.79 You do not have to take my word for it. I share the rules, code, and even a free backtester on my blog. You can inspect the logic and test it yourself. The Resource Room → Interactive Models → Buying Short-Term Dips in Stocks

How do I know whether a trading system is broken, or just going through an unfavorable market? One check I find useful: compare its equity curve with those of similar models. For me, this starts with trading simple models whose logic I understand. Not a pile of data mined rules whose only justification is a nice backtest. Over time, I usually find other people trading the same basic idea. Their equity curves become useful reference. The chart shows the live trading performance of Ares, tracked on Collective2. I do not know the developer, but the system appears to trade NQ intraday breakouts, similar to what I trade. It trades more frequently, but the overall behavior looks very familiar. Neither equity curve has been making new highs lately. That does not mean a system is broken. Sometimes the market favors your approach. Sometimes you have to be patient. What would concern me is a comparable model consistently making new highs while mine kept struggling. That would make me question whether my rules were too closely fitted to the past, rather than simply blaming the market. I use the same exercise for my other trading styles too.

“Buying 0DTE options won’t work. You’re fighting theta, and trading costs will eat you alive.” That was some of the feedback I received. So I opened a fresh Interactive Brokers account dedicated to one fully automated strategy, just to track its live results separately. It follows my intraday volatility breakout logic, but executes through long 0DTE options. Blue is the strategy account’s performance. Green is the SPY benchmark. I would not trade this as my only strategy. The equity curve is too volatile for my taste. But it does not tie up capital very often. Losing trades are frequent, but the winners can be large. That is how I use it at a larger scale in my main account: as one of several ways to execute and manage my intraday breakout trades - not something the entire portfolio depends on. The live results do not mean theta and costs are irrelevant. But they are a useful counterexample to the blanket claim that “it will not work.”

I keep getting requests to review new “AI-powered” analytics, backtesting, and trading bot platforms. Yes, AI makes these tools much easier to build. But that is exactly why I am skeptical about the business model. The same technology that helps you build the product also helps your potential customers build their own version. Tailored to how they actually trade. If another trader can recreate the core functionality in a few weeks, what will keep them paying you every month? Not everyone wants to build their own software. But making something easy to build does not automatically create demand for another subscription. I suspect the market for these small, generic trading SaaS products is much smaller than their builders expect. Building tools for yourself or your trading team? I see enormous value in that. Solve your own problems. Improve your workflow. Spend more time on the actual trading and research. But do not confuse “I can build this” with “other traders will subscribe to this.”

A very interesting project to check out for anyone exploring systematic trend following and momentum in futures. It builds heavily on Robert Carver’s work, but with some different implementation choices. What I particularly like is the dashboard. It makes a complex portfolio easier to follow, showing orders, daily P&L drawdowns etc. There is also an ongoing paper trading run, with daily updates published to GitHub. So you can study the implementation and follow how the portfolio behaves as new data arrives. Worth exploring for ideas, even if your own approach is different. github.com/Lucas-Joly-GH/tre…

I see less value in trading tools as AI makes it easier to build what I need. What is difficult to build today will likely be much easier in the future. What becomes more and more valuable is the data. Collect as much relevant data as you can, even if you do not know how to analyze it yet. For Interactive Brokers users a practical example is regularly downloading Activity Flex reports. Your trading journal may preserve some of that information, but not all of it. With the right sections selected, Flex reports can capture much more than trades: commissions, other fees, interest, FX activity, deposits and withdrawals. So I would store the data in a local database and keep the original export files. Later, an LLM can help build detailed reports from that history. You do not need to have every report or every question figured out today. But you do need the underlying data. Without your own archive, you are limited to IBKR’s available Flex history. AI can help me rebuild a reporting tool. It cannot recover records that are no longer available.

Even a simple trend following system on individual stocks can better than passive buy and hold, in my view. The one I trade is straightforward: • Select a stock universe for the month. For example using momentum. • Trade long Donchian breakouts within that universe throughout the month. • Manage exits with an ATR trailing stop. • Include gold through GLD to diversify beyond stocks. Not a holy grail. But a useful addition to my other stock systems. It also leaves capital available that I can put to work elsewhere in the portfolio. The ATR trailing stop gets me out relatively quickly when a move reverses. The biggest potential improvement I see is in universe selection. Breakouts with trailing stops work well on stocks making strong, sustained moves. Currently, I use a very simple momentum based approach to select the monthly basket. But I am interested in testing whether AI could help identify stocks that are “hot” for fundamental reasons.

A simple trading system can still be heavily overfit. That is why I do not judge a system only by its final backtest. I also want to know how it was developed. "Only two parameters" tells me what ended up in the model. It tells me nothing about how many alternatives were tested. Different markets. Different timeframes. Different entry conditions. Different exits. Test enough combinations and show only the best version, and the final rules can still look deceptively simple. I like simplicity because it makes a system easier to understand, test, benchmark and maintain. But I also want to know which decisions were made before seeing the results - and which were made because of them. The result matters. But without understanding the research process, I do not know how much to trust it.

The best feedback on a strategy post is not likes or reposts. It is an email one year later: “Hi Peter, Just wanted to send my thanks to you for the QQQ Z-Score strategy that you shared on X about a year ago. I have actually been using it live since then and it has performed extremely well!” I shared the full rules and code here about a year ago. The chart shows the strategy’s continuously updated backtest. But that message shows something more valuable: a trader could understand the logic, test it independently, and put it into live practice. That is exactly why I publish my research. Sometimes one useful idea can improve how someone researches, tests, or trades for years. I have benefited from other traders’ ideas in exactly the same way. That is why I keep sharing mine.

Mean reversion systems are usually associated with fast entries and short holding periods. This one is deliberately slow. It is an RSI based model built on a weekly timeframe and applied to S&P 500 constituents, with many years of out-of-sample history. I would not allocate an entire account to it. Its CAGR is not especially attractive relative to the average capital it employs. But Inside a broader portfolio, it can provide useful diversification through slower signals, longer holding periods, and different behavior from faster mean reversion systems. I publish its current open positions on my blog under: Live Trading Models → Buy the Dip (Weekly) A system does not need to be exceptional on its own to improve a portfolio.

Not all volatility is useful to a breakout trader. 2026 has produced plenty of intraday movement in US indices, but much of it has reverted instead of developing into sustained trends. From what I see, intraday breakout models trading almost every session have had a difficult year: many breakouts, little follow-through, and repeated losses. My model is more selective. It trades only on sessions with a historically higher probability of becoming trend days. That has meant fewer trades this year. And I am glad about that. A filter is not supposed to maximize activity. It is supposed to reduce exposure when conditions are less favorable. On QQQ, my main market, the equity curve is still moving in the right direction. The edge is not only in the breakout entry. It is also in knowing when not to participate.

The #1 mistake I see traders make is not choosing the wrong system. It is risking too much on a system that looked convincing. The sizing feels reasonable while the equity curve is smooth. Then the first real drawdown arrives, uncertainty rises, and the trader either overrides the rules or stops completely. Lower returns may feel disappointing. An intolerable drawdown can end a trading career. I am comfortable starting with an imperfect system. I am not comfortable starting with reckless sizing. Better conservative than naive.