ATMResearch
Join Overfit - free

StatOasis · by Ali Casey

Stop guessing. Start building.

Research, build, test, combine, deploy: the loop that turns trader confidence into conviction.

Get the best trading research in your inbox.

Running a quick security check before this can be sent.

Delivered +2 times a month - when the work is ready, not on a calendar.I'll never sell your address.Unsubscribe in one click, any time.

Turn knowledge into process.|

Happy members

Is this you?

live trading beginsIn-sampletraders' backtestOut-of-sampletraders' realitypeaklive trading beginsIn-sampletraders' backtestOut-of-sampletraders' realityIn-sampletraders' backtestOut-of-sampletraders' realitylossgain
Illustrative: a clean backtest is a hypothesis, not proof.
  • 5,800+read OverFitnewsletter
  • 3,500+members inthe community
  • 23,800+subscriberson YouTube
  • 300+ATM studentsglobally
  • 440+videoson YouTube
  • 30+Q&A video sessionsin the community
  • 40+deep diveresearch articles
  • 5,800+read OverFitnewsletter
  • 3,500+members inthe community
  • 23,800+subscriberson YouTube
  • 300+ATM studentsglobally
  • 440+videoson YouTube
  • 30+Q&A video sessionsin the community
  • 40+deep diveresearch articles

The loop most traders skip three steps of.

Five stages, running as a loop - every live strategy feeds the next round of research. The red shortcut is the one most traders take instead.

Most traders follow this path and call it done.
1

Research

Start from a behavioral hypothesis, not a prediction.

2

Build

Turn the hypothesis into rules, not a curve fit.

3

Robustness test

Out-of-sample, walk-forward, across regimes the optimizer never saw.

4

Portfolio construction

Combine strategies that fail at different times.

5

Live deployment

Size it, monitor it, plan the drawdown before it comes.

Most traders follow this path and call it done.

  1. 1

    Research

    Start from a behavioral hypothesis, not a prediction.

  2. 2

    Build

    Turn the hypothesis into rules, not a curve fit.

    ↷ Most traders follow this path and call it done.

  3. 3

    Robustness test

    Out-of-sample, walk-forward, across regimes the optimizer never saw.

  4. 4

    Portfolio construction

    Combine strategies that fail at different times.

  5. 5

    Live deployment

    Size it, monitor it, plan the drawdown before it comes.

back to research

Can you survive a decade of real markets?

A decade of markets. Five decisions. About eight minutes. See how your instincts hold up against the regimes that broke real strategies, then see the trading personality your choices reveal.

Start: Survive the Decade

I'm Ali Casey.

I build systematic trading strategies and teach the workflow behind them. Not a content creator. A system builder who happens to teach.

More about me →
Portrait of Ali Casey

Happy Members

5.0
3.5K+

Algo trader since 2014

Here is how I can help you.

Overfit - the newsletter logoOverfit - the newsletter logo

Overfit - the newsletter

Skip the hype. Trust the data. One practical, evidence-driven takeaway per issue: strategy testing, portfolio construction, market structure, and the mechanics behind systematic trading - every claim with the backtest behind it. Publishing since 2024, formerly The AlgoTrader.

Learn more→
Algo Trading Masterclass logoAlgo Trading Masterclass logo

Algo Trading Masterclass

The complete build-a-system workflow, taught end to end: finding an edge, testing it honestly, assembling strategies into a portfolio, and putting it live - the same process behind every study I publish. Currently open as a waitlist.

Learn more→

Community

Where builders share strategies, code, and feedback. Get plugins, correlation book, indicators and more. Free to join today.

Learn more→

Digital Products

Everything you can download and use the same day: books, strategy packs, and tools - starting with 36 Ways to Buy the Dip, backtested mean-reversion entries for systematic dip-buying, every variant tested, every result shown. Listed on the products page with everything else StatOasis makes.

Learn more→
AlgoChef logoAlgoChef logo

AlgoChef

The validation layer I built for the step most traders skip: pressure-testing a strategy before it risks real money. Launched 2026.

Learn more→

Latest from StatOasis YouTube channel

Nasdaq Fell 5.5% Thursday. I Tested the Day After.

Nasdaq Fell 5.5% Thursday. I Tested the Day After.

Nov 9, 2026

Two Trading Indicators, One Identical Trade

Two Trading Indicators, One Identical Trade

Nov 2, 2026

Donchian Channel Strategy: 283 of 300 Made Money. Why?

Donchian Channel Strategy: 283 of 300 Made Money. Why?

Oct 26, 2026

Latest from StatOasis X channel

StatOasis card. One-time licence prices. Three editions. One real difference. $1,290 Starter; $1,490 Professional; $2,900 Ultimate, the standout in orange; $300 in blue as a different kind of number, Extend support after year one.

A review ranking above mine lists StrategyQuant X Professional at ~$2,490 and Ultimate at $4,900. I read the vendor's page on 19 Aug 2026: $1,490 and $2,900. It quoted the struck-through price as the price. Check the date on any price you read. https://statoasis.com/overfit/research/strategyquant-x-review-the-no-code-way-to-build-trading-strategies

Nov 20, 2026 · 50 impressions
StatOasis card. 131 made money. 3 beat random. Donchian trend long on 7 stock index markets: 131 of 140 variants made money, 3 of 140 beat a random entry per bar held, and 111 of 900 variants beat random overall where chance gives about 45.

I ran 900 Donchian channel backtests on 15 markets, 2004 to 2026. Breakout and fade, 5 channel lengths, long and short. Every variant was set beside 100 random entries with the same exit. Which side paid, and did the breakout beat chance?

Nov 19, 2026 · 84 impressions · 2 reactions
Card titled "The most widely recommended filter finished last of eight." A large 28.1 percent, the share of 96 matched configurations where volume above the 50-day average improved profit factor, the fewest of any volume filter tested. A callout states its opposite, volume below 0.8 times average, improved profit factor in 71.9 percent of the same pairings, the widest gap between any filter and its mirror in the study.

"Add a volume filter to your strategy." I tested 80 of them on Larry Connors' RSI(2). 15,552 backtests, SPY from 1993 to 2026, S&P 500 futures from 2007 to 2026. The filter everyone recommends finished last of eight. 🧵

Nov 18, 2026 · 84 impressions · 2 reactions

Latest from LinkedIn

Card titled "A Sharpe screen is precise. It is also brutal." A development Sharpe of at least 0.5 scored 94.5 percent precision out of sample, passed 16.7 percent of candidates and rejected 75.2 percent of eventual survivors. A callout says the parameter-neighbour check rejects only 23.5 percent of them.

A high in-sample Sharpe is the most precise validation check I have tested. It is also the most wasteful one that works. I scored 11 checks on 5,472 strategies. Each was computed on the first 75% of history and graded on the final 25%, which none of them saw. Screen on a development Sharpe of at least 0.5: 94.5% of the strategies it passed were still profitable out of sample. It passed 16.7% of candidates. It rejected 75.2% of the strategies that would have survived. Three of every four good strategies died in the screen. The parameter-neighbour check (are the settings either side of yours also profitable?) scored 91.7% and rejected only 23.5% of the eventual survivors. Under three points less precise, at a third of the cost. Every check has a price, paid in strategies you would have been happy to trade. Most checklists never show that column. What the study measured is the screen. Ranking the survivors by Sharpe is my practice, not something this study tested. Full table, with the cost of every check: https://statoasis.com/overfit/research/is-your-backtest-overfit-validation-checklist

Nov 20, 2026
StatOasis card. With the drift: 283 of 300. Against it: 43. Donchian trend rule variants that made money by side on 15 markets: 283 of 300 with the drift, 43 of 300 against it, 0 of 140 short on stock indices. With the drift is known only after the fact.

900 Donchian channel backtests. The side mattered more than any setting. A Donchian channel is the highest high and the lowest low of the last N bars. The classic trade buys a close above yesterday's upper line and shorts a close below yesterday's lower line. I ran that breakout on 15 markets from 2004 to 2026: channel lengths from 10 to 200 bars, four exits, long and short. Traded with each market's drift, 283 of 300 trend variants made money. Against it, 43 of 300. "With the drift" means long where buy and hold made money over the window, and short where it lost. On the 7 stock index markets the split was total: long 131 of 140, short 0 of 140. Read that grouping for what it is. It uses each market's whole 22-year result, which nobody had in 2004. It is known only after the fact. It tells you where the money came from. It is not a rule you could have traded. And making money is not the same as beating chance. Set beside random entries with the same side and exit, 111 of 900 variants beat the random 95th percentile per bar held. Chance alone gives about 45. Every table, all 15 markets: https://statoasis.com/overfit/research/donchian-channel

Nov 19, 2026
StatOasis card. The filter, swept 139 ways. Best cell on one market. Worse than nothing on five. 1 of 139 Rank of the published pair on ES; 137 of 139 Its rank on IWM; 6 of 6 Markets where the median filter loses; 0 of 139 Pairs beating no filter on DIA.

Adding a filter to a strategy almost always improves the backtest. That is the problem, not the feature. Here is a worked example on a weekly mean reversion rule, tested 22,680 ways across six index markets. The filter is Volume Oscillator (95,100) above zero, the gap between a 95-week and a 100-week average of traded volume. Instead of trusting that setting, I held the rule completely still and swept only those two numbers: 139 period pairs, from 5 weeks to 150, scored by the same code on all six markets. On E-mini S&P 500 futures, 95/100 ranked 1st of 139. First, out of every setting available. On the other five markets the same setting ranked 62nd, 56th, 92nd, 39th and 137th. On 5 of the 6, using no filter at all beat it. On the Dow ETF, not one of the 139 settings beat leaving the filter off. Ninety-five weeks against one hundred weeks is a five-week difference measured across two years of volume. Move either average one step on the sweep's grid, or both, and ES return-to-drawdown lands anywhere from 1.60 to 7.75. What there is, is a search that ran over a lot of settings and reported its winner. The rule underneath survives, for the record: buy the week after a down week, sell that week's close, no filter. Profitable on all six markets, with a smaller worst drawdown than buy-and-hold on all six, 54.4% to 58.2% win rates, about 22 decisions a year. The test costs nothing. Before you trust a filter, sweep its settings. If it pays on one market and fails on the others, you have found a search artefact, not an edge. Full audit: https://statoasis.com/overfit/research/a-simple-weekly-mean-reversion-strategy-for-index-futures

Nov 18, 2026

What you won't find here.

No trade calls. No daily predictions. No PnL screenshots. This is for traders who want to understand how to build trading systems, how they behave in different market regimes, when they break, and why. If that's not you, there are plenty of other places, and no hard feelings.

What traders say.

rodolfo berrocal's photo

rodolfo berrocal

Hello, my name is Rodolfo Berrocal, I'm from Lima, Peru. I just stopped by to thank you for all the great work you do. I'm doing my best to understand your videos since I don't speak English very well, but I'm still learning a lot from you. I'm just getting started in the world of algorithmic trading, and it's going well for me. Thank you again, and blessings.

Tommy's photo

Tommy

Dear Ali, your attention to detail - and people - was always bar none. No question, you are our Financial Guru, while also a loving/caring figure so you deserve our utmost respect! Thank you for all your shared experience and everyday hard work for our not-only-financial wellfare! :)

Chad's photo

Chad

I’ve been in the program since Feb, have found a way to create and find a lot of pretty good algos, but magic sauce happens when we combine the uncorrelated “pretty goods”. Happy to report since Feb, up 28pct and am now in maintenance mode working on the process instead of reflexively reacting to every draw down that I think is the end of the world. Biggest learnings. Test for worst case portfolio drawdown, size accordingly. Trade top strategies within that portfolio, cycle out worst performers every few months. Seems to be a winning recipe and am so excited to think about the next chapter in this journey I’ve been on for many many moons 😃

Read more →

StatOasis is calm, evidence-based algorithmic-trading education, founded by Ali Casey. Ali builds systematic trading strategies and teaches the workflow behind them: research, build, test, combine, deploy. He writes the Overfit newsletter, published since 2024, and runs the Algo Trading Masterclass.

Socials

  • X ↗
  • YouTube ↗
  • Instagram ↗
  • LinkedIn ↗
  • GitHub ↗
  • Muck Rack ↗
  • LinkedIn SO ↗
  • GitHub SO ↗

Products

  • Overfit - the newsletter
  • Algo Trading Masterclass
  • StatOasis Community
  • 36 Ways to Buy the Dip
  • AlgoChef ↗

Reading & tools

  • Research
  • Methodology
  • Survive the Decade
  • Wall of Love

StatOasis

  • About Ali Casey
  • Contact
© 2026 StatOasis. Calm, evidence-based.
PrivacyTermsHypothetical resultsCalifornia