For Links To All Strategies Click Here

ATS publishes automated trading strategies (C#). Some of those strategies are in the forward test. The Weekly Alert selects which of those strategies in the forward test (and Incubator) look like the strongest candidates for the coming week, produced by a model called V1.5. The following week, every call gets reviewed against actual P&L. Bad weeks get reported the same way good ones do. That is the point of a public model.

The goal: use the V1.5 Weekly Model to select strategies for automated portfolio managers (APMs) to follow. To learn more about APMs and where we are in the testing process, click here.

In this week’s Weekly Alert we will:

• Score last week’s picks— both versions

• Share 5 picks for Apr 27 to May 1

• Intro a new model coming May 3

• Highlight the macro week ahead: FOMC, PCE, GDP, four Mag 7 reports

So how did the model perform last week:

Read more

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

Well that’s a wrap on Wave 2. Time to talk lessons learned, and what is changing for Wave 3, which starts on May 10.

If you are new here, I am obsessed with portfolio management/strategy selection. To answer that obsession I am building automated portfolio managers, or APMs. An APM sits on top of a portfolio of automated trading strategies and decides which trades to actually take based on an internal logic. For this Wave 2 Experiment, three APMs (Hadiya v24, v25, and v26) are sitting on the same ten child strategies, each governing its own simulated account. You can read more about it by downloading the Wave 2 guide, which can be found here.

So how did it go? In this post we’re going to look at:

• what the child portfolio did;

• how Hadiya v24, v25, and v26 performed;

• what Week 3 exposed;

• why Selene v1 is joining Wave 3.

By the end of this post, I think you’ll agree that we are on the brink of something big. It’s the feeling of seeing tracks before seeing the animal.

Let’s get into it…

Here is the per-account weekly P&L for the full three-week run. This is based on running 1 contract of each instrument:

The children gave back $19K in Week 3. S3 on GC bled almost exactly $11,000 every single week.. The only strategy in the cohort that never had a green week. There is something almost noble about it.

Worth noting on the other side of the trade: Crude was the second-best child strategy this month. If you traded crude in April 2026 and you did NOT make money, write down what you were doing and stop doing that.

So what about our APMs: Hadiya v24, v25 and v26? How did they perform? Did they outperform the children? Will we have a repeat of W1 where child strategies underperformed, but all three PMs still landed positive? Or did the PMs crash?

Read more

For Links To All Strategies Click Here

4 of 5 calls were hits for the second week in a row on this V1.5 model!

Net P&L: Up +$49,648 across the five calls this week. If you had followed last week’s alert and the one before it, that is 8 hits on 10 calls and +$90,228 cumulative, which is either a model finding its footing or the setup for a very educational third week. Here’s a look at the equity curve.

The $30K hole was entirely house money. It was profit earned Monday, partially given back Tuesday, then re-earned by Friday. Max drawdown is the standard metric used to describe this hole, but it is not the whole picture. The same curve looks very different depending on which definition you use.

Two weeks, 10 calls, 8 hits, +$90K net against a max drawdown of roughly $31K. Roughly three dollars made for every dollar underwater. Three-to-one return over drawdown is a respectable ratio.

Calls For 4-19 through 4-24

Before getting into this week’s calls, I want to share an answer to a question that someone asked me last week. They asked, “what is all of this about; what’s the end goal?”

That is an evolving answer, but here’s my elevator pitch…

I am in the process of developing two parallel approaches to strategy selection with the hope that they may both feed into the best live executions. The first is automated: a family of portfolio managers (APMs) that manage live signals in real time. You can read more about that here. The second is this weekly alert, an AI assisted model that scores every strategy in the forward test and selects the highest probability candidates for the coming day/week.

Each week the Weekly Alert is scored against actual results.

So how’s the project coming along?

For the second week in a row, we’re at 80 percent for the weekly alerts. If you had followed last week’s alert and the one before it, that is 8 hits on 10 calls or $90,228. What about the APMs? The flagship APM (Hadiya v25) is the clear winner over the last two weeks in the Wave 2 Experiment, producing $73K while the underlying portfolio produced $56K. The best part is that the APM did this with 58% less drawdown.

The APM is getting these result from the first three strategies ATS published (Strategies 1, 2 and 3). They were chosen for simplicity more so than anything else. What will happen when we use the Weekly Alert strategies as the underlying portfolio for the APMs? Fun times ahead!

We’re not there yet, but I wanted to share the vision as it stands today.

Now let’s get into the picks for this week…

Read more

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

If you’re new here, I am obsessed with strategy selection. To answer that call I’m building automated portfolio managers, or APMs. An APM sits on a portfolio of strategies and makes decisions about which trades to take. For this Experiment, it is three portfolio managers that govern ten child strategies. Each APM trades its own account.

The child strategies, parameters, data series and instruments used for each strategy are included in the Wave 2 Reset Guide. I tell you everything you need to get set up there and if you have any issues, or get stuck, send me an email.

In addition to strategy selection, APMs also solve a question I get all the time: what happens when you want to run a portfolio of strategies on one brokerage account, ie, one child strategy is long NQ and another is short NQ at the same time?

This is the conflicting-position problem, which becomes a portfolio allocation problem. Portfolio allocation is about deciding how much capital sits behind each strategy. If you have $100K and ten strategies, you’re implicitly (or explicitly) deciding how much margin and risk each strategy is allowed to consume. Does Strategy 3 on GC get the same allocation as Strategy 1 on HG? Are you sizing by margin, by expected return, by risk contribution, or trade duration (my personal favorite)?

For APM’s, the allocation problem is solved with logic and netting in two ways:

First, allocation numbers become fictional when strategies net each other out. If you’ve allocated 10% to Strategy A (long NQ) and 10% to Strategy B (short NQ), and they cancel, your actual NQ exposure is 0%. The allocation framework says you’re deployed, but the reality is you’re flat.

Second, your effective allocation defaults to whatever the netting math produces, not what you designed. When two strategies conflict, the one that entered first determines the position. The second one either reduces it or flips it, depending on size. So your actual portfolio composition is shaped by signal timing rather than by your allocation decisions.

This is where APMs come in. An APM is making an allocation decision at every single signal. It is baked into the selection logic.

So how is the Wave 2 Experiment going?

We’ve got three PMs running side by side: Hadiya v24, v25, and v26. Each one manages the same ten child strategies using a slightly different selection logic.

Week 1 result: all three PMs finished green on a week where the raw child portfolio lost $7,424. That is the result I am looking for. Here’s how the child strategies performed that first week:

And this is how the three APMs performed against those child strategies:

v25 was the clear winner. APM Hadiya v25 performed better than the child strategies and the two other Hadiya APMs.

Where are we after two weeks? Is APM Hadiya v25 still performing better than the other two APMs?

First let’s look at how each of the child strategies performed over the last two weeks combined:

Note: Be sure to use the data series I give you in the Wave 2 Reset Guide to get these results.

This was a great week for child strategies. Especially Strategy 1 and 3 (NQ). The raw child portfolio put up $63,149 on 210 trades. Friday alone contributed $24,385. The best PM (v24) managed $43,171 on 191 trades. v25 came in at $43,000, and v26 managed only $14,509.

And this is how each PM performed over the same two week period:

v25 is still the clear winner over the two week period producing $72,628 while the underlying portfolio produced $55,725. The best part is that APM Hadiya v25 did this with 58% less drawdown.

What exactly does that mean?

Let’s take a minute to review a few drawdown definitions to make sure we’re on the same page:

• Max Drawdown (Max DD): The largest peak-to-trough decline in the equity curve. If equity goes from +$50K to +$23K, that is a $27K drawdown. It measures the worst fall from your best point, the deepest hole you had to sit through.

• Absolute Drawdown: How far the account dropped below its starting value. If you started at $0 and equity hit -$7K before recovering, absolute drawdown is $7K. This answers: did I go underwater, and by how much?

• Active Drawdown: Where you are right now relative to your equity peak. If your peak was +$50K and you are currently at +$40K, active DD is $10K. It is the live version of drawdown. Are you in a hole right now, and how deep?

These numbers together tell you something net P&L alone just can’t: how much pain did it take to get here?

As you can see from the chart below, the children strategies made $55,725, but to get there, they endured a $65,099 max drawdown, meaning at one point, the equity curve was $65K below its peak. In contrast, v25 made more money ($72,628) with less than half the drawdown ($27,258).

And, its Net/DD ratio is 2.7, which means that for every dollar of max pain, we get $2.70 in profit. That is 3x more capital-efficient than child strategies alone.

v25 did not just beat the children on net. It beat them on every risk metric.

Here’s an equity curve to help further illustrate the difference in drawdown. As you can see, the child strategies experience the greatest trough at -$65K.

So the next obvious question is: what is the logic for each of these APMs and what’s so different about v25?

Read more

For Links To All Strategies Click Here

ATS is developing two parallel approaches to strategy selection. The first is automated: a family of portfolio managers (APMs) that filter live signals in real time. The second is this weekly alert, an AI assisted model that scores every strategy in the forward test and selects the highest probability candidates for the coming week.

Each week the Weekly Alert is scored against actual results. The goal is not to predict the market. It is to identify which strategies are most likely to be net positive over the next five trading days.

Why does the economic calendar matter to a strategy selection model? Because strategies are not indifferent to what is happening in the world. A mean reversion system that thrives in quiet sessions can get destroyed by a surprise PPI print. A momentum strategy might love the directional follow through after bank earnings. The model does not try to predict what the numbers will be. It uses the calendar to understand what kind of week is coming, and which strategies are best suited to that kind of week.

Last week’s Weekly Alert Calls produced the best result yet.

The Scorecard: 4 of 5 Calls Were Hits

Hit Rate: 80 percent, the best to date

Net P&L: +$40,580 across the five calls

Read more

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

As you all know, I am obsessed with strategy selection. In addition to building a weekly alert system, we’re also in the process of building automated portfolio managers or APMs. An APM sits on a portfolio of strategies and makes decisions about which trades to take.

The first experiment was a failure. Wave 2 resulted in a reset due to bad logic related to connection issues. Last week we started over with the Wave 2 Reset— all fixes were applied.

How did it go?

All three PMs ended Week 1 in the green on a week where the underlying ten-strategy portfolio lost $7,424. Here is how the 10 child strategies performed last week:

Here is the Week 1 APM comparison.

All three versions beat the child total portfolio, and they did it on a week where the underlying portfolio lost money. That is the result we’re looking for. The loss queue, the clustering detection, the Trap state, the exposure caps, the risk modulation— all of that machinery exists to answer one question: does an automated portfolio manager have the ability to outperform the underlying portfolio? After one week, the answer is yes for all three versions, and the spread between them is telling us which selection logic works best.

What’s the difference in selection logic?

All three APMs share the same logic, with a slight variation. v25 was by far the winning logic system. So what’s the difference between each system and what’s so special about 25?

Read more

This was a strong week. The market gave us clear directional moves on Monday and Thursday. 71% of accounts were profitable. The biggest winners were NQ strategies on longer timeframes, while Strategy106 on micros delivered steady volume.

Every strategy below finished the week with over $100 in net profit. These strategies are from the Q2 Forward Test and Incubator

For Links To All Strategies Click Here

• SimStrategy12 NQ: +$23,690 on 12 trades. 58% win rate. PF 3.51. Top performer by a wide margin.

• SimStrategy3 GC: +$16,470 on 30 trades. 40% win rate but large winners pulled it through. PF 1.67.

• SimStrategy3 NQ: +$15,105 on 24 trades. 42% win rate. PF 1.80. Consistent presence in the top three.

• SupportResistance NQ: +$14,665 on 6 trades. Only 2 of 6 won, but the winners were large. PF 3.40.

• SupportResistance NQ: +$10,680 on 2 trades. Both winners.

• OptimalLegs HighWin NQ: +$9,330 on 10 trades. 60% win rate. PF 4.73. Excellent risk-adjusted returns.

• OptimalLegs Balanced NQ: +$9,135 on 11 trades. PF 4.62. The OptimalLegs pair continues to deliver.

• Strategy77 NQ: +$9,110 on 2 trades. PF 12.61. Small sample but massive.

• Strategy9 NQ: +$7,050 on 5 trades. 60% win rate. PF 1.88.

• Strategy3 NQ: +$5,795 on 11 trades. 55% win rate. PF 1.61.

• Strategy106 MES: +$4,578 on 145 trades. The highest-volume profitable strategy this week. PF 1.45.

• Strategy106 M2K: +$3,366 on 255 trades. Highest trade count of any strategy. 54% win rate. PF 1.61.

• SimAccount89 GC: +$3,150 on 220 trades. 62% win rate. Max MAE of $10,010. This strategy takes heat but comes back. PF 1.03.

• SOCBasedIntraday NQ: +$2,920 on 9 trades. PF 1.55.

• FundedTrader NQ: +$2,175 on 7 trades. PF 1.19.

• Strategy41 ES: +$2,000 on 2 trades. PF 7.40.

• SimAccount0 GC: +$1,910 on 17 trades. 88% win rate. One of the highest win rates in the portfolio.

• SimAccount0 NQ: +$1,365 on 34 trades. 82% win rate.

• Strategy106 MCL: +$813 on 262 trades. Micro Crude. PF 1.06. Grinding.

• Strategy106 MGC: +$353 on 198 trades. Micro Gold. Barely positive. PF 1.02.

Note: I am not forward testing all published strategies. Some have been forward tested, others have not. If you’ve found a high performing variation, let me know and I’ll add it to the list. I will be publishing the two OptimalLegs strategies in the coming months.

For Links To All Strategies Click Here

I am currently obsessed with strategy selection. We’ve found strategies with edge, but I want to enhance that edge; I want to leverage allocation in a way that maximizes both risk and opportunity.

At the moment, ATS runs two parallel approaches to strategy selection.

• The first is automated: a family of portfolio managers (APMs) that filter live signals in real time.

• The second is this weekly alert; an AI assisted model that scores every strategy in the forward test and selects the highest-probability candidates for the coming week.

This alert model is a work-in-progress. Each week it is scored against actual results. The goal is not to predict the market. It is to identify which strategies are most likely to be net positive over the next five trading days.

Why the economic calendar matters. Strategies are not indifferent to what is happening in the world. A mean-reversion system that thrives in quiet sessions can get destroyed by a surprise CPI print. A momentum strategy might love the directional follow-through after FOMC. The model does not try to predict what the numbers will be. It uses the calendar to understand what kind of week is coming. Is it event-dense or quiet, are the releases clustered on one day or spread across the week, are there Fed speakers who could rock the market? That context shapes strategy behavior.

Let’s take a look at the economic week ahead.

• Monday, April 6: No top-tier US macro release. The week starts clean. Use this day to position based on Friday’s ISM prints and the NFP number.

• Tuesday, April 7: Rescheduled Data and Fed Talk

  • 8:30 AM ET: Durable Goods Orders (Forecast +0.6%). This is the February report, rescheduled from its original March 25 date due to the government shutdown, so the data is stale but the market may still react sharply.

  • 3:00 PM ET: Consumer Credit. Usually a minor note, but worth watching for signs of tightening.

  • 5:50 PM ET: Fed Vice Chair Jefferson speaks on “Economic Outlook and the Labor Market.” Late-day volatility sleeper for those running overnight positions.

• Wednesday, April 8: The FOMC Reveal

  • 2:00 PM ET: FOMC Minutes from the March 17-18 meeting. This is where we look for the why behind the Fed’s rate path. Expect a high-velocity window between 2:00 and 3:00 PM as algorithms parse the text.

• Thursday, April 9: The Data Dump

  • 8:30 AM ET: Five major releases hitting the tape simultaneously at the open:

    • Core PCE (the Fed’s preferred inflation gauge)

    • GDP Q4 Third Estimate

    • Personal Income

    • Personal Spending

    • Initial Jobless Claims

• Friday, April 10: CPI and Consumer Sentiment

  • 8:30 AM ET: Consumer Price Index. If Thursday’s PCE is the anchor, Friday’s CPI is the confirmation or contradiction. This is not a quiet day.

  • 10:00 AM ET: Michigan Consumer Sentiment (Preliminary). Consumer confidence in the current tariff and rate environment.

No NFP this week. The Thursday-Friday PCE-CPI sequence is what matters.

With that context, let’s see what last week’s call score was.

What gets scored

Each weekly call is an enable decision: this specific strategy, on this specific instrument, for the coming week. Every call is scored the following week as a hit or a miss based on whether the strategy was net positive. The goal is not to predict the market. It is to identify which strategies are most likely to be net positive over the next five trading days.

Last week’s calls scored 2 out of 3.

SimAccount89 GC hit (+$3,150). SimStrategy3 NQ hit (+$15,105). SimStrategy89 NQ missed (-$8,695).

That miss led to a minor upgrade in model logic that I’ll share with you in a moment. First, let’s take a look at the calls for the coming week.

Scored Calls For The Week Ending April 10

The model is making five calls for this week. That means you want to enable all five of these strategies.

Read more

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

This post is for those of you participating in the Wave 2 experiment for APMs. It’s not too late to join because we’re starting over.

If you are new here, a quick orientation. ATS runs a three-tier testing system: the Incubator (live data on simulated accounts), the Forward Test (live data on simulated account), and the Live Test (real funded accounts). Automated Portfolio Managers, or APMs, sit on top of individual strategies and decide in real time which trades to copy and which to filter. They are the layer between having edge and knowing when to use it. Wave 2 is a head-to-head test of three versions of our primary APM, Hadiya, to answer how that filtering should be scoped.

A number of you have come to me with great questions over the last two weeks. The test group also had a big finding that in hindsight should have been most obvious, so we need to reset.

The original Wave 2 started on March 22 with a clean signal pipeline and three versions of Hadiya running head to head across ten strategy slots and eight instruments. The contamination problem from Wave 1 was solved. The blocked entry tag filter, the anti-cascade prefix filter, and the exit signal filter were all in place. The experiment was finally asking the right question: which version of the Automated Portfolio Manager Hadiya is best?

Then NinjaTrader 8 reminded us it has a mind of its own.

I love NinjaTrader 8, but it is fragile. Connections drop. The platform crashes. When it comes back online, it does not always come back cleanly. Over the first two weeks of Wave 2, even with the use of a VPS, this was happening more often than expected, and the result was duplicate trades.

Read more

Hellooooo ATS,

I wanted to take a moment at the start of Q2 to do two things. First, to say thank you. Second, to tell you where this is going.

Where we have been.

The last five months have been the most productive stretch in the history of this publication. If you have been here for all of it, you know. If you joined recently, here’s a quick overview of s…

Read more

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

We are back. Contract roll week is behind us. What you are reading today is the first issue under the new framework. It is simpler, more honest about its limitations, and every call within it can be scored.

Let’s get started…

One-Minute Overview

This is an event-dense week. Powell speaks Monday; JOLTS and Consumer Confidence land Tuesday; Wednesday delivers a “triple stack” (ADP, Retail Sales, and ISM); and Nonfarm Payrolls (NFP) anchor Friday.

Read more

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

If you study markets for long enough, you will begin to see structures. Trends, ranges, pennants, head and shoulders, double tops, double bottoms, wedges, channels, consolidations, breakouts, pullbacks, V-reversals, rounded bottoms, cup and handle. Every one of these appears on a 1-minute chart, a 15-minute chart, a daily chart, and a weekly chart. The pattern does not care about the time scale. And the structures nest inside each other like a fractal. What we see depends entirely on the resolution we choose to look at. The bar frequency is that resolution.

There is always structure within structure. Chop inside trends inside ranges inside larger trends. The question we’re answering with this indicator is: at what scale does the current structure live, and does my strategy match the structure at the scale I am trading?

Imagine two traders are talking after hours:

“What data series are you running that on?”

“1000 tick.”

“Why 1000?”

“Backtest said so.”

A bond trader overhears this and walks over.

“What maturity are you buying?”

“10 year.”

“Why 10 year?”

The bond trader would never say “backtest said so.” He would point at the yield curve.

The yield curve tells him the term structure of interest rates across every maturity. It tells him where the compensation is, where the risk is, and where the regime shifts are. The shape of the curve is the answer to “why this maturity?”

A steep curve means one thing. An inverted curve means something else. A flat curve means something else again. Bond traders do not pick a maturity and hope it works. They read the curve and allocate accordingly.

The futures trader has no equivalent. She has a backtest, and the backtest is built on a foundation that is riddled with errors. The tick data has gaps, bad prints, and vendor-specific quirks that vary by provider and by year. The backtest engine introduces its own distortions: fill assumptions that do not match live execution, bar-building logic that handles edge cases differently than real-time, and optimization routines that reward overfitting.

The most consequential parameter of the entire setup: bar frequency, which determines what the market looks like to the strategy.

Bond traders have the yield curve as a native map of structure across maturities. Futures traders have no universally adopted equivalent for structure across bar frequencies, which is part of what makes this indicator interesting.

That is the gap. The Bar Frequency Curve Indicator I’m going to share with you today fills that gap.

This adds a new dimension to strategy selection (my current obsession). Instead of locking a strategy to one bar frequency and hoping the structure cooperates, we can scan across frequencies in real time and deploy at the one where the frequency supports the logic. The strategy goes where the signal is.

I think this will meaningfully increase the number of trades that have structural support behind them, and decrease the ones that do not, which should theoretically increase win rate.

The Connection

I have always suspected there was a connection between bar frequency and price behavior. Not a vague connection. A fundamental one. The kind where if I could measure it, I could use it to put a strategy in resonance with the market rather than waiting for a static market to match a certain condition set.

We treat bar frequency as a fixed decision, and that can work at higher bar frequencies and/or highly robust strategies, but what if the market’s behavior is defined by the frequency?

That would change the game. Instead of building a strategy and then hoping the market cooperates, we could look at the market’s structure across frequencies and select the one where our strategy has structural support. We would not be figuring out how to survive in the dimension we are stuck in. We would not be waiting until conditions happen to align with the dimension we chose from a flawed backtest. We would be selecting the best dimension to operate in, right now, based on what the market is actually doing at each scale.

Think of it like tuning a radio. A radio does not wait for the station to come to its frequency. It scans across the dial until it finds resonance, a clear signal. A strategy should do the same. Scan across bar frequencies. Find where the structure is clear. Operate there.

Right now, NQ might be mean-reverting at 100 ticks and trending at 750 ticks. Both are market structures. Both are real. They coexist because the market is fractal in nature. It is not in one regime. It is in multiple regimes simultaneously, one at each scale. The Bar Frequency Curve shows you which one lives where.

Read more

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

Well, I’ll just come right out and say it—the forward test had unplanned gaps this week. The exact cause is not confirmed. Any call made for Wednesday daytime is unscoreable because no strategies were active during that session. This will be noted in the scorecard accordingly.

I will work to ensure continuous VPS operation going forward. Where gaps occur they will be documented explicitly so the scorecard reflects only windows where the test was actually running.

All anomaly designations in this alert are derived from triple-lens validation. A designation earns Structural Dead Zone (SDZ) or Structural Edge Window (SEW) status only when the per-trade signal is consistent in direction and magnitude across all three lenses with adequate volume in each. Single-lens or two-lens findings are Watch Items only.

All designations govern strategy entry enablement only. Exit logic is handled by each strategy independently.

Equity contract roll week is a work week for ATS, so no report next week. All tests are disabled this week, but I’m still providing a scorecard and this week’s calls. Additionally, there are certain strategies that always perform well on Triple Witching and I will tell you what they are as well.

Here’s a one minute overview of top alerts for the week ahead…

Read more

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

One of the hardest aspects of automated trading is strategy selection. I am hoping that the weekly anomaly model will greatly assist in this effort and I look forward to the day when the scorecard can boast 90%+ accuracy.

I have also created automated portfolio managers (APMs) to assist with this effort.

An APM is a meta strategy that runs above your strategies and decides which signals to act on and which to ignore. Rather than letting every strategy fire freely, an APM watches execution patterns across the portfolio and filters trades based on what the market is actually doing.

The question was never just whether a strategy had edge. The question was when to let it trade. APMs are the answer to that question. APM Hadiya is our current best attempt at one.

Who Is APM Hadiya?

Wave 1 of the APM Experiment ran 46 automated portfolio managers head to head for 33 days across 120,173 trades. The core failure was contamination: between 40 and 63 percent of all trades were NinjaTrader housekeeping events being misread as entry signals. That problem is now solved along with a dozen or so other issues dealing with phantom trades.

Not all APMs failed. Six survived profitable. APM Hadiya was the most profitable.

Hadiya is a loss-pattern detection system built on a Hidden Markov Model with four hidden states. Wave 2 runs three versions of the APM simultaneously.

A small group of you have been in the trenches with me on this review process. It has been demanding work, and I do not take that lightly. On behalf of everyone who will benefit from what gets built here, thank you.

Based on their feedback, three versions of Hadiya have been created. Wave 2 runs three versions simultaneously. Each one has the same logic, but deploys in a different way. The full setup guide below provides detail on the difference between v24, v25 and v26.

Wave 2 starts clean on March 22, the week after contract roll. The goal is to ask the question: which version of the Hadiya brain performs best at filtering live signals across multiple instruments and asset classes?

The full setup guide is available for download below. It covers account creation, contract months, roll schedule, APM configuration, and what to watch in the Output window. If you are running alongside me, that is your starting point.

What We Hope to Answer

Running the same setup across multiple subscriber machines is not just about sharing results. It is about finding out whether the mechanics hold. Does the anti-cascade filter behave the same way on every NT8 instance? Do the filter decisions match across different environments running identical configurations? If two people running the same setup get different results, that is a mechanics problem, not a strategy problem. The mass experiment is the stress test.

Beyond that, four research questions drive the HMM side of the experiment.

• Does group-based HMM avoid the over-filtering problem that portfolio-level HMM creates?

• Does auto-calibration in v26 outperform the fixed threshold in v25?

• Which asset group pair produces the most useful HMM signal contrast?

• Does selectivity beat volume?

The APM that filters most aggressively should, in theory, produce the best risk-adjusted return. Wave 1 suggested this was true. Wave 2 is a much cleaner test. Wave 3 is where the real fun happens. We’ll apply fixes learned via APM Haidya to additional automated portfolio management systems like cross momentum, conformal prediction and entropy.

Fun times!!!

Questions? Reply to this post or reach out directly.

Automated Trading Strategies: AutomatedTradingStrategies@protonmail.com

Read more

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

One of the hardest aspects of automated trading is strategy selection. I am hoping that the automated portfolio managers some of you are helping me to build will greatly assist in this effort and I look forward to sharing the details of the Wave 2 Experiment with you shortly. The experiment was pushed to a start date of March 22, the week after contract roll.

I have also created a weekly anomaly model for strategy alerts. The current model is a work in process, but it’s getting better. It selects strong candidates based on historical performance, current events and anomalies. These are the alerts for March 8-13, 2026.

The framework you are reading today looks different from the one I published in January (or even last week), because ten weeks of real forward test data told me where it was wrong. That improvement process never stops.

The original framework had three components: time-of-day anomalies derived from forward test data, week-of-month patterns, and event-day guidance around major economic releases. It was a reasonable first draft. It was also making calls the data did not fully support, applying portfolio-level warnings to instruments that did not need them, and had no language for positive windows at all. It was more avoidance map than strategy alerts. The new framework should also help with accuracy.

The metric I use to measure and improve accuracy is defined as the percentage of published calls that hit or partially hit, weighted and tracked cumulatively. Right now the live score sits at 56.25%, but I’ve made a lot of changes to the model this week. When I retroactively applied today's updated framework to all ten weeks going back to January 4, the score came out at approximately 91%. That number is not a live result. It reflects what the model would have called if the current rules had been in place from the start.

What is an anomaly?

An anomaly, in the way I use it here, is a statistically recurring deviation from what random market behavior would predict. These are not one-off surprises. They are patterns that keep showing up in the forward test data when they theoretically should not if markets were perfectly efficient.

Every anomaly is derived empirically from my own forward test data. They are filtered through strategies I actually run on live data using simulated accounts.

One critical clarification on methodology: every time-of-day designation in this framework governs strategy enablement for entry, not exit management. The dead zones and edge windows describe when strategies should be enabled to enter a trade. Once a strategy is in a trade, it runs its own exit logic.

Before getting into the alerts for this week, I owe you a scorecard on how well last week performed.

Scorecard: March 1-7, 2026

Read more

For links to all strategies click here

As a quick reminder, we’re on the hunt for the holy grail of automated trading strategy. If you have any questions, start with the FAQs and if you still have questions, feel free to reach out to me (Celan) directly at AutomatedTradingStrategies@protonmail.com.

I do not usually publish things like this. The ATS focus is trading, not macro commentary, but the February jobs report pulled me out of that lane, and not for the reason you might expect. The number itself, a 92K miss, was not the surprise. What was surreal was watching the entire financial establishment scramble to explain it without once mentioning the very thing that a different set of very smart people warn us about every single day.

The architects of AI cannot stop telling us that this technology is going to reshape the economy at a scale we are not ready for. Then 92,000 jobs vanish and the explanation is a snowstorm? Really? The juxtaposition is too strange to leave alone.

So here are the numbers that came out this morning:

• Non-Farm Payrolls: -92,000 (Consensus: +59,000)

• Unemployment Rate: 4.4% (Consensus: 4.3%)

• Labor Force Participation: 62.0% (Lowest since December 2021 outside the pandemic)

The actual print came in 151,000 below consensus. It’s the kind of gap that, in any recent year, would have been considered shocking on its own. The March 2019 miss, widely remembered as a market-moving disaster, was a 160,000 deviation. We are in that territory folks.

Yet these were the headlines:

• CNBC: “The U.S. economy lost jobs in February, a month marred by severe winter weather and a strike at a major health-care provider.”

• CNN (preview, March 5th): “Severe winter storms likely put a big damper on hiring in weather-sensitive industries like construction and leisure and hospitality.”

• CNN: “Economists were expecting job growth to slow — in part due to a major labor strike by health care workers and a deep cold snap that hit many US states.”

• Fox Business: “There are a handful of things that may have distorted February’s data: winter storms may explain the weakness in construction.”

What do these headlines all have in common—they all blame the snow.

The specific storm being cited is Winter Storm Hernando. Hernando formed and struck February 22 through February 24. That is eight days after the survey period closed. The BLS was not measuring employment during Hernando. It was measuring employment during a survey week that preceded Hernando by more than a week. The February payroll number is tied to the pay period including Feb. 12, so the Feb. 22–24 storm was mostly not in the payroll count; at most it could affect a limited slice of employers with longer pay periods, and weather generally hits hours more than headcount.

So Wall Street is quite literally using a weather event that happened eight days after the survey closed to explain why 92,000 workers vanished?

So What’s Really Happening?

What actually happened during the survey week is confirmed by the BLS release itself. The Kaiser Permanente strike sidelined more than 30,000 healthcare workers in California and Hawaii during the reference period, which subtracted 28,000 from payrolls. The strike has since been resolved. That is a legitimate, temporary, and documented drag. The release also says that even adjusting for the strike, payrolls would have been deeply negative.

The weather story is not just wrong on the chronology. I believe it is serving a function. If the miss is weather, analysts/business leaders do not have to admit their models are missing a structural shift in hiring behavior. A temporary distortion means you buy the dip. A structural breakdown means revaluation. The incentive to blame the snow is real even when the snow story is demonstrably weak.

The structural damage to the labor market has been building for two years and is now large enough to measure clearly if you look at the right data.

The metrics that actually show where we are paint a consistent picture. The quits rate sits at 2.0%, below its 2019 pre-pandemic average. Workers quit voluntarily when they have somewhere better to go. They do not quit when they feel trapped. The hires rate is running at levels last seen in 2013, when the economy was still recovering from the financial crisis. Job openings fell to 6.5 million in December, with the vacancy rate at 3.9%, the lowest in nearly a decade outside the pandemic. There were almost one million more unemployed people than available jobs at year-end, the widest such gap since 2017.

If all of this is simply cyclical, a strike here, some cold weather there, then why are the people building the technology behaving the way they are?

In the past month alone, three high-profile departures from the top AI labs made news for the same underlying reason. Mrinank Sharma, who led the Safeguards Research Team at Anthropic, resigned on February 9th. Zoe Hitzig, a researcher who spent two years at OpenAI resigned the day OpenAI began testing advertisements inside ChatGPT. Hieu Pham, an engineer who had worked at both OpenAI and xAI, stepped away from frontier AI entirely and relocated his family to Vietnam. The “fleeing the machine before it collapses” energy is difficult to dismiss when it comes in this concentration and with this consistency.

Then there’s AI business and thought leaders…

In January, Dario Amodei, the CEO of Anthropic (Claude), warned of a “shock to the job market bigger than any before.” He previously stated AI could wipe out half of all entry-level white-collar jobs within five years. In February, Sam Altman, CEO OpenAI, said that “The real impact of AI doing jobs in the next few years will begin to be palpable.” In December 2025, Geoffrey Hinton, the “Godfather of AI”, said that “One of the main sources of money is going to be by selling people AI that will do the work of workers much cheaper.” He called 2026 the year contracts for human labor simply stop being renewed, which is exactly what seems to be happening, disproportionately for workers under 25, driven not by terminations, but by a hiring door that stopped opening.

Information services lost 11,000 jobs in February as part of a 12-month trend averaging 5,000 losses per month. That sector is being directly shaped by AI-related cuts, and the market knows it. Block announced it would slash roughly 40% of its payroll in direct response to AI capabilities. IBM made similar moves. These are not isolated events.

Marc Benioff, CEO of Salesforce, said AI is already doing 50% of his company’s workload. Andy Jassy, CEO of Amazon, said corporate headcount will shrink because of AI. He told employees to prepare to “get more done with scrappier teams.” Jim Farley, CEO of Ford, said AI will “replace literally half of all white-collar workers.” Doug McMillon, CEO of Walmart, told the WSJ that AI “is going to change literally every job.”

The contrast is incredible.

You have the literal architects and billionaires building the technology screaming from the rooftops that the white-collar labor market is about to evaporate, and meanwhile, CNBC and Jefferies are on television blaming a snowstorm.

The Trading Setup

I wouldn’t be a trader if I wasn’t looking for the trading angle, so here goes….

Every leading indicator is deteriorating. Openings are at decade lows. Hires are at 2013 recession-recovery levels. Quits are below pre-pandemic norms. Long-term unemployment has been above the historical danger threshold of 25% of total unemployed since mid-2025. Participation is falling. The only metric holding is weekly claims, and that is the tripwire. The trigger is weekly claims above 250,000 sustained for three to four consecutive weeks. Everything else is already in place. Claims is the confirmation.

Equities are priced for a soft landing the labor data no longer supports. The cleanest expression of that mismatch is to short equity exposure or go long on volatility with a sustained claims spike. The rates trade is even murkier because inflation limits the full safe-haven bond bid.

The slow bleed has already added up for the people living it, but the risk here, as is the case for most trades, is timing. I believe this low-fire dynamic can persist longer than logic suggests it should, especially with the media behind it. In situations like this, position sizing matters more than being right on direction.

Questions? Reply to this post or reach out directly. Automated Trading Strategies automatedtradingstrategies.substack.com

Coming soon:

• APM, Live Test and Forward Test Update

• Strategy Alerts For March 8 - 13th

• Post: How To Select The Best Strategies To Trade For The Day

• Post: Strategy 109: The Ice Fishing Strategy

• Post: Awareness Exercises For Traders

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

This is another big week. Get your screens ready!

One of the hardest aspects of automated trading is strategy selection. I am hoping that the automated portfolio managers (APMs) some of you are helping me to build will greatly assist in this effort. I have also created a weekly anomaly model for strategy alerts. The current model is a work in process. It selects strong candidates/alerts based on historical performance and current events/anomalies. These are the alerts for March 1-6, 2026.

An anomaly, in the way I use it here, is a statistically recurring deviation from what random market behavior would predict. These are not one-off surprises; they are patterns that keep showing up in the forward test data when they theoretically should not be if markets were perfectly efficient.

I track several categories:

• Time-of-day anomalies are intraday patterns baked into market microstructure.

• Week-of-month anomalies are patterns tied to specific weeks of the calendar month.

• Event-driven anomalies are behavioral signatures the market leaves around scheduled economic releases: NFP, ISM, ADP.

Every anomaly is derived empirically from my own Incubator and Forward Test data. They are filtered through strategies I actually run on live data using simulated accounts.

This week I am introducing a triple-lens validation. I am cross-referencing every anomaly against three independent datasets: the Q1 2026 Forward Test, the Historical Forward Test going back to January 2024, and the Incubator. An anomaly that shows up in only one dataset gets flagged. An anomaly that shows up in all three earns the highest conviction.

Before getting into this week, however, I owe you a scorecard on how well last week performed.

How Did Last Week’s Forecasts Perform?

The alert model is only useful if I track what it gets right, what it gets wrong, and why. Here is the accounting for last week, February 22-28.

Read more

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

This is a BIG WEEK! Get your screens ready!

As you all know, I’ve been testing a weekly model for alerts. The current model selects strong candidates based on historical performance and current events/anomalies.

An anomaly, in the way I use it here, is a statistically recurring deviation from what random market behavior would predict. These aren’t one-off surprises; they are patterns that keep showing up in the forward test data when they theoretically shouldn’t be if markets were perfectly efficient.

I track several categories:

• Time-of-day anomalies are intraday patterns baked into market microstructure. The 2 AM cluster is the flagship example this quarter: NQ 03-26 averages plus $341 per trade at 2 AM across 1,235 trades. These are the most persistent anomalies because they’re driven by structural mechanics: thin order books, institutional positioning flows, and the absence of retail noise.

• Week-of-month anomalies are patterns tied to specific weeks of the calendar month. Week 2 is the only positive week in the full forward test dataset across all 151,626 trades. This likely reflects institutional rebalancing flows, options expiration cycles, and systematic fund activity that clusters around predictable calendar dates. These are more persistent than day-of-week patterns because they have an institutional anchor.

• Event-driven anomalies are behavioral signatures the market leaves around scheduled economic releases; NFP, CPI, FOMC. I’m not predicting the number. I’m trading the shape of the market’s reaction: suppressed volatility ahead, whipsaw during, directional resolution after. The post-announcement 10 AM window is the most underexploited finding in the current dataset.

Every anomaly is derived empirically from my own Incubator and Forward Test data. They’re filtered through strategies I actually run.

One important caveat: not all anomalies are created equal. Some patterns are structural and rock-solid. Others are regime-dependent and shift year to year or even week to week.

Before getting into the weekly strategy alerts for February 22–28, 2026, we need to discuss the elephant in the room because it has the potential to rock the markets next week.

Read more

For links to all strategies click here

As a quick reminder, we’re on the hunt for the holy grail of automated trading strategies. If you have any questions, start with the FAQs and if you still have questions, feel free to reach out to me (Celan) directly at AutomatedTradingStrategies@protonmail.com.

Damn It Feels Good To Be A Trader

Here’s some comic relief from one of my favorite movies…

The Q1 Forward Test is in full stride, and what a quarter it’s turning out to be. If you’re a trader, you’re probably having one of the best stretches in recent memory.

Why? I could speculate and waste our time, but the underlying effect is undeniable: greater volatility and higher intraday swings, especially on the NQ.

We’ve got two portfolios this quarter, both relatively small compared to previous quarters. I’ll show you the exact strategies in each test in a moment.

Here’s a look at the equity curve for both:

In total, FT1 is up $138K on 3,654 trades. FT2 is up $220K on 1,721 trades.

Both forward test portfolios operate side by side on the same data, same time period — January 4 through February 7, 2026. FT1 is the original: 10 strategies across NQ, GC, HG, and ZB, generating 3,654 trades. It’s well-diversified, near-zero correlated, and textbook in its construction. FT2 is the challenger: 9 strategies, more concentrated, fewer trades, built around higher-conviction setups.

The early numbers favor FT2:

FT2 trades 53% less and makes 60% more. Average trade jumped from $38 to $128. Max drawdown got shallower. Sharpe went from 4.15 to 6.33.

But look at the last two rows. FT1 wins on correlation, and that’s not a small thing. The question I’m trying to answer over the rest of Q1 is whether that diversification advantage shows up when it matters most: during sustained drawdowns and regime changes.

Here’s where it gets interesting.

FT1 is a beautifully diversified portfolio. Average pairwise correlation of 0.05 — basically zero. 42% of all strategy pairs are negatively correlated. Even as instrument relationships in the broader market break down, ZB (bonds) hedged against GC (gold) at -0.41. MNQ offset NQ at -0.28. Textbook portfolio construction.

FT2 runs hotter. Average correlation is 0.12 with 38% negative pairs. The reason is NQ concentration — some pairs correlate as high as 0.73. Metal strategies provide balance, but it’s a fundamentally more concentrated bet.

So the real question becomes: is that extra diversification worth the drag?

Right now, FT2’s higher edge per trade appears to compensate for its concentration risk. Max drawdown actually decreased 22% despite the higher correlation. You’d expect a more correlated portfolio to draw down harder. When one strategy bleeds, correlated strategies should bleed together.

So why didn’t it?

Because drawdown isn’t just about how many strategies lose at the same time. It’s about how much each strategy loses relative to how much it earns. FT1’s mediocre strategies were contributing to drawdown without contributing proportionally to recovery.

And here’s the real discovery: FT2’s strategies recover faster because they earn more per trade. A strategy making $1,028/trade claws back a $5,000 loss in 5 trades. A strategy making $13/trade needs 385. When your winners are 3.5x larger on a per-trade basis, time spent in drawdown shrinks, even if the drawdowns are correlated. Higher expectancy = faster recovery = shallower peak-to-trough.

So what makes FT2 produce fewer trades and higher profit? What in the logic is driving this?

We’ll get into that in a moment because it’s a very interesting story, and it clearly points to a certain type of strategy logic that thrives in this environment.

Before I get into the logic, let me show you the actual strategies in both portfolios.

Forward Test 1 and Forward Test 2 Portfolio Composition

The chart below shows strategy, trade count, net profit, and profit factor for each strategy in both portfolios. Forward Test 1 is on the left, Forward Test 2 on the right, and the core (shared) strategies are in the middle. Click here for links to all ATS strategies.

Read more

For links to all strategies click here

As a quick reminder, we’re on the hunt for the holy grail of automated trading strategies. If you have any questions, start with the FAQs and if you still have questions, feel free to reach out to me (Celan) directly at AutomatedTradingStrategies@protonmail.com.

Just a quick update on three things:

• Q1 2026 Forward Test

• Live Test #2

• APM Battle

Q1 2026 Forward Test

The Forwrad Test started last week. This is the smallest portfolio I’ve ever run. It consists of approximately 7 strategies.

Here’s a look at the equity curve:

In total, the portfolio is up $46K on 661 trades. Here’s a daily breakdown:

Now let’s talk about the Live Test.

Live Test #2

Live Test #2 started last week as well. To read more about the first Live Test (196% return), click here.

No trades were made last week. Analysis from 2025 showed that Week 2 is the best performance week of the month and it appears to be fairly persistent. The first week is very important because it determines your drawdown. Best case scenario, this week will fund the rest (read: future drawdowns won’t ever dip below Week 1 profit).

The Great APM Race

An APM is a meta-strategy that sits on top of your underlying trading strategies and makes real-time decisions on each trade signal.

How it works: Your base strategies (Strategy 3, Strategy 6, Strategy 9, etc.) generate trade signals. Instead of blindly executing every signal, an APM evaluates each one against its specific criteria. Each APM uses a different methodology. A strategy might have a 1.2 profit factor trading every signal. But if an APM can identify which 40% of signals are high-quality and filter out the rest, that subset might have a 2.0+ profit factor. You’re trading less, but making more.

The best part (at least to me) is that these APMs aren’t static filters. They’re designed to learn and adapt. As they accumulate more trade history, their models recalibrate. Cecil refines his prediction. Tomas updates his state probabilities. Usha adjusts her weights. The expectation is that performance should improve as each APM builds a larger sample of what works and what doesn’t in current market conditions.

How long until we see meaningful learning? Honestly, I don’t know. It could take weeks to accumulate enough data for statistical significance, or months before the adaptive mechanisms show clear improvement over baseline. Some APMs may converge quickly while others need more runway. This is uncharted territory.

Currently, Tabitha (Tobacco Lite), is the only APM I’ve published. She finished Week 1 at ($8,075), which actually makes her the 2nd best performer in the Tobacco family, behind only Tomas (V5), which was up $13K. I will share the V5 and V6 APMs with you the first week in February. Meanwhile, if you haven’t done so already, start creating your forward test portfolio. You need at least 10 strategies that trade on a regular basis for this to work. You can run the same strategy, ie, Strategy 3, on 10 different instruments.

Week 1 Winners:

• 🥇 Cecil: +$17,332

• 🥈 Tomas: +$13,394

• 🥉 Maryam: +$10,082

Here’s a bar race showing the race from day-to-day:

This is the first time I’m running 34 APMs head-to-head in live forward testing. And I still have about 10 that have yet to make a trade.

This little experiment could be a complete failure or we might be on the heels of discovering a truly novel approach to portfolio management. I know as much as you do at this point. We’re watching this experiment unfold together in real-time.

Week 1 gave us Cecil in the lead, but one week proves nothing. The real story will emerge over months of data. That’s what makes this interesting. We’re not backtesting hypotheticals, we’re running a live tournament and letting the results speak for themselves.

If you have any questions feel free to reach out directly:
Contact AutomatedTradingStrategies@protonmail.com

I will be away from my desk until February so response times might be delayed. All subscribers have been comped for one month.

Subscribers, look for a special gift (Strategy 77 and 83c) from me on Christmas Day!

I’m on a quest for the holy grail of automated trading. Questions? Check the FAQs or feel free to reach out directly: AutomatedTradingStrategies@protonmail.com.

For Links To All Strategies Click Here

The holidays are here! But the best part of the holidays isn’t the gifts. It’s the table. Not a desk. Not a screen. A proper table. Set with intention and surrounded by people you don’t see often enough. A table weighted with dishes that only appear once a year and food that takes longer to prepare than it takes to eat.

There’s something about a shared meal that resists optimization. You don’t rank the courses by caloric efficiency. You don’t A/B test the wine pairings. A great meal isn’t a collection of great dishes—it’s magicical. Acid responds to fat. Salt awakens sweet. Wine does the rest. And when it works, frowns ease into smiles, shoulders drop, and the table becomes the reason everyone came.

I’ve been thinking about this as I build the Q1 2026 Forward Test portfolio.

200+ strategies in the Incubator. Our pantry is overflowing, but a pantry isn’t a feast until someone decides what belongs on the table.

Welcome to À La ATS.

This post is about composition. What belongs on the table together? What gets left in the kitchen?

The Pantry Problem

200+ strategies. Data from September 19th to present—a full quarter of forward testing through one of the choppier Q4s in recent memory. Trend systems and mean reversion. Tick-based microstructure plays and multi-hour session trades on NQ, ES and GC. HMM regime detection layered onto classical setups. Time-based edges that ignore price entirely.

I have an abundance of strategies to choose from, but which ones will make it to the Q1 2026 Forward Test? The temptation is to sort by net profit, pick the top ten, size up, and be out. It’s the temptation I’ve had for the last 8 quarters and I’ve followed it, but here’s what else I’ve learned: a leaderboard of profitable strategies isn’t a portfolio. Seven versions of the best potato salad isn’t a feast.

I’ve been asking: Which strategies made the most money and/or had the highest profit factor.

Perhaps a better question would be: Which strategies can I trust with capital together?

Suddenly we’re looking at structure, not rankings. I’m focusing on the whole, not the part. I should be looking for strategies that grind, explode, fail at different times—which means they protect each other. Perhaps, instead of a buffet—pile your plate, hope for the best—I should approach this as a five-course meal.

Mise en Place: The Table Setting

In professional kitchens, mise en place means everything in its place. Onions diced. Stocks reduced. Proteins tempered. Before a single pan heats up, the work is done.

At a formal dinner, place settings serve the same function. The oyster fork sits outside the salad fork because you’ll need it first. Wine glasses descend in size from water to white to red—the order they’ll be filled. Every element has a reason. Nothing is arbitrary.

Portfolio construction requires its own mise en place:

• Max contracts — No single position dominates the plate

• Max daily loss — Remove the dish before it spoils the meal

• Max consecutive losses — Pause between courses; reassess the pace

• Session filter — Some ingredients only arrive fresh at specific hours

This is a Portfolio Constitution. The rules that hold even when the kitchen gets hot.

The Main Course: Boeuf Bourguignon

Boeuf Bourguignon doesn’t impress with flash. It impresses with depth. Beef braised for hours in red wine, onions, mushrooms, bacon. The kind of dish that feeds a table and leaves everyone quiet with satisfaction.

The main course of a portfolio works the same way. These are strategies that grind, preserve capital, absorb noise. They don’t make headlines. They make money.

What qualifies for the center plate?

• Trade count high enough to trust (100+ minimum)

• Moderate profit factor but consistent (>1.5)

• Average MAE controlled relative to the instrument

• Max MAE not wildly larger than average (no hidden gristle)

From the Incubator:

• Strategy3 (NQ) - The beef itself. Reliable, flavorful, does the work.

  • Net profit: $89,555

  • Trades: 307

  • PF: 1.53

  • Avg MAE: $933 / Max MAE: $7,620

  • Textbook core-engine behavior. Smooth participation, reasonable drawdowns, no heroics.

• SimAccount3 (ES) — The potatoes. Nobody talks about them, everyone eats them.

  • Trades: 324

  • PF: 1.52

  • Avg MAE: $460

  • Instrument diversification plus lower MAE equals portfolio ballast.

• Strategy 104 _ DualHMM (NQ) — The wine reduction. Concentrated, elevated, transforms everything it touches.

  • PF: 2.31

  • Avg MAE: $659 / Max MAE: $1,770

  • High profit factor plus low MAE. Rare. This is the best risk-adjusted performer in the Incubator.

• SOCBasedIntraday (NQ) —The pearl onions. Small, precise, essential texture.

  • Trades: 115

  • PF: 1.83

  • Shorter-term structure-based edge. Shines when the slower systems stall.

These four form the center; the main course. They sustain.

What about dessert (aka: the convex sleeve). These are some of my favorites:

• Supportresistancetrend (NQ) — The soufflé. Rises spectacularly when conditions are right. Sits flat otherwise.

  • PF: 3.5–4.0

  • Trade count: Low

• Timey_67 (NQ) — Crème brûlée. Crack the surface, find the gold underneath.

  • PF: 3.08

  • Avg MAE: $478

• ChannelMeanReversion (NQ) — Baked Alaska. Fire and ice. Works precisely because it shouldn’t. Mean reversion at the microstructure level. When trend systems chop, this one hunts different prey.

The Shape of Edge: What To Put On The Table

As much as I like to use stats, when it comes to selecting the best strategies for this new portfolio design, I want to start placing more focus on the shape of the curve. I’ll be sharing the equity curve for 30 profitable strategies from the Incubator with you in a moment. The curves serve as a bird’s-eye view of the strategy’s payoff structure.

As you will see, the shape of each curve tends to fit three profiles: linear, convex and concave.

Linear Payoff:

• What it is: Profit and loss move proportionally with the market. Risk $1, make $1. Lose $1, down $1. No leverage, no asymmetry.

• Strategy examples: Buy and hold. Simple trend-following without stops. Delta-one positions.

• Character: Balanced exposure. You participate equally in ups and downs.

Convex Payoff (Lose Small, Win Big)

• What it is: The curve bends upward. Upside is larger than the downside. Risk a little to potentially make a lot. The payoff accelerates as you get more ‘right’.

• Strategy examples: Buying options. Trend strategies with tight stops and runners. Breakout systems. High profit factor / low trade count strategies.

• Character: Most trades lose or scratch. But winners are multiples of losers. You’re paying a small premium repeatedly for occasional big wins.

The Tarte Tatin was invented by accident at the Hotel Tatin in 1880s France. Stéphanie Tatin overcooked apples in butter and sugar, panicked, threw pastry on top, flipped it, and served it anyway. It became one of the most celebrated desserts in French cuisine.

Convex strategies work the same way. They’re not about expected value. They’re about payoff shape—small losses most of the time, then a regime arrives and they deliver something memorable.

You don’t build a meal around dessert. But a feast without one is…forgettable.

Read more