What I Do Before I Even Consider a Stop Loss,Take Profit or Entry

Most traders pick a stop loss based on a feeling. They look at the chart, find a level that looks right, and put their stop there. Then they wonder why they keep getting stopped out. I run six mathematical steps before I think about where to put a stop. These steps tell me what kind of edge I have, how that edge behaves in different market conditions, and whether I should take the trade at all. All before a single dollar of risk is defined. This is the measurement phase of my pipeline.

The Pipeline

The full pipeline has eight steps. Steps one and two handle data cleansing and strategy coding in Python and NinjaTrader. That's the foundation. Today we're starting at step three: the fixed constant. We'll cover every measurement step that happens before you ever set a stop loss.

Bottom line up front. Three things you'll take away from this article.

1. Fixed constants give you a consistent anchor point. Same level, every day, for everyone.

2. MAE and MFE distributions tell you what price actually does from that level.

3. Markov classification tells you what kind of edge you have, and how to trade it.

Step 3: The Fixed Constant

Before we measure anything, we need something to measure from. In this pipeline, we call that a fixed constant.

A fixed constant is a price level that meets three criteria. It's deterministic: same data in, same number out, every time. It's known in advance: the level is set before you need to make a decision. It's time-bound: you know exactly when it's established.

Example: the high and low of the first five-minute candle at 10 AM. Every trading day, at 10:05 AM, those two numbers are locked. 21,350 is the high. 21,320 is the low. Everyone looking at the same chart sees the same levels. No interpretation. Other examples include yesterday's high, the VWAP at 10 AM, the opening print at 9:30, the Asia session range. All fixed constants.

Why It Matters

When we measure MAE and MFE (how far price moves against you and for you), we need a consistent anchor point. If you measure from a support level that two traders draw differently, your measurements fall apart. One trader's MAE comes out at 0.10% and another's at 0.25% because they drew the level at a different spot. With a fixed constant, every trader gets the same measurement.

Can you measure MAE and MFE from a pattern? A double bottom or an order block? Technically yes, but patterns are discretionary. Two traders see different ones; patterns don't occur every day either. You might get 30 instances in a year. P95 of 30 trades is literally the second-largest value in the set. That's an anecdote, not data.

A fixed constant occurs every trading day. Over five years of trading, that's 1,250 instances of the same level forming. That's not a guess; that's a distribution. Step three of the pipeline asks one question: what is your fixed constant? Everything after that is math.

Step 4: The Raw Measurement

Once we have the fixed constant, step four measures what price does from that level. No stop loss. No take profit. No filters. No trade management. We're not trading here. We're observing.

For every instance of the level forming (every 10 AM box high across 1,250 trading days over five years), we track two things. MAE: maximum adverse excursion, how far price moves against you from the level before the trade resolves. MFE: maximum favorable excursion, how far price moves for you.

Three Hold Periods

We measure at three hold periods: one hour after the level forms, two hours, and three hours. The same level might produce a strong one-hour move that reverses at two hours. Or it might chop for an hour then trend for three. You need the full picture, not a single snapshot.

Three Lookback Windows

Five years gives you the structural view. 1,250 measurements across crashes, rallies, bear markets, recovery. This tells you what is possible. Your catastrophe stop should survive this distribution.

One year gives you the current view. About 250 measurements. Markets change; the 2022 environment where everything was crashing looks nothing like 2025's sustained trend. Your working stop and take profit should be calibrated here. What's happening now, not three years ago.

Ninety days is the freshness check. About 60 measurements, not enough for reliable percentiles. This window is a warning system. If the 90-day MAE median runs 30% higher than the one-year median across those same measurements, something changed. Volatility is expanding.

The Percentile Table

From those measurements we build percentile tables. P10 means ten percent of trades had MAE less than this value. P50 is the median. P90 means ninety percent of trades stayed below this level; only ten percent went further. Across 1,250 trades over five years, the median MAE sits at 0.10% and median MFE at 0.14%. At P90, MAE is 0.32% and MFE is 0.38%.

Write this down. This table is your strategy. The stop loss, the take profit, the cover-the-queen area, the position size: all derived from these numbers. Not from a feeling. Not from a chart level you eyeballed. From 1,250 measurements of what price actually does.

Step 5: The Markov Chain

Step five takes the measurement data and asks the most important question: does price trend away from your level, or chop around it? This is the Markov chain classification.

We look at how MAE and MFE grow across the three hold periods. In one example across 1,250 trades: MAE grew from 0.10 to 0.16 (1.6x growth). MFE grew from 0.14 to 0.20 (1.43x growth). Rho equals MAE growth divided by MFE growth: 1.6 / 1.43 = 1.12. Rho greater than one means MAE is growing faster than MFE. That's a win-rate edge.

Win-rate edge means you win more often than you lose, but the wins are small. If you slap a take-profit target on a win-rate trade, you cut your winners short. The edge lives in holding to the close.

Different example. MAE grew 1.3x. MFE grew 2.0x. Rho = 0.65. MFE is outpacing MAE; price is trending in your favor. That's a momentum trade. Here you want a take-profit target, something around MFE P50 or P60.

Why Rho Changes Everything

This one number determines your entire exit strategy. Before you look at a stop loss, you need to know whether you're trading momentum or win rate. The stop, the target, the cover-the-queen level, the hold time: all different depending on the answer.

We check rho across all three lookback windows. If five-year rho says momentum but 90-day rho is drifting toward 1.0, the edge is changing. Maybe the market structure shifted. Maybe this level doesn't trend like it used to.

Write this down. Check rho across all three windows. If 90-day rho drifts toward 1.0, the edge is degrading.

Step 6: The Regime-Switching Model

Step six asks a question most traders never consider: does the edge behave the same in calm markets and volatile markets? Almost always, the answer is no.

We use VVIX, the volatility of the VIX. The market's fear gauge for the fear gauge. When VVIX runs high, options traders are panicking about volatility itself. We compute the median VVIX over the full five-year dataset; days above the median count as volatile, days below count as calm. We also track the daily range ratio (today's NQ range divided by the ten-day median range) and VVIX rate of change.

The Regime Flip

We take all 1,250 trade measurements from step four and split them by regime. Every trade has a date. We look up VVIX on that date. Two separate tables come out.

Calm days, about 500 trades out of the 1,250: MAE P50 is 0.08%, MFE P50 is 0.14%, rho = 0.67. Strong momentum. Volatile days, about 750 trades: MAE P50 is 0.15%, MFE P50 is 0.18%, rho = 1.20. Win-rate edge.

Read that again. Same level. Same strategy. Same everything. In calm markets it's a momentum trade; in volatile markets it's a win-rate trade. Nine out of 24 configs in our box strategy (37% of all configurations tested across 1,250 trades each) flip classification between regimes. If you're not checking for this, you're running the wrong playbook more than a third of the time.

Hold Period by Regime

In calm markets, rho might stay below 1.0 at one hour, two hours, and three hours. The momentum holds; you can sit in the trade longer. In volatile markets, rho might be below 1.0 at one hour but flip above 1.0 at two hours. The edge degrades after the first hour.

Same level. Different market. Different hold time. This is why we measure everything three times. Not to cherry-pick. To adapt.

Step 7: The Hidden Markov Model

Steps four through six are structural. They tell you what kind of edge you have and when it changes. They don't tell you whether to take this specific trade on this specific day. Step seven does.

The Hidden Markov Model looks at 30 observable signals before each trade and predicts whether conditions are favorable or unfavorable. The states we're trying to detect (good conditions vs. bad conditions) aren't directly observable. No single metric tells you a trade will be a winner. But the combination of 30 signals creates a pattern the model can learn.

The 30 Features

Market conditions: 2 features. Sequence: 2. History: 3. MAE hit rates: 10, covering what percentage of trades hit each percentile level in the trailing year versus the trailing quarter. MFE reach rates: 10, same structure. Streak context: 3. Thirty total.

We train using the Baum-Welch algorithm, a standard ML technique that discovers hidden states from observed data. BIC (Bayesian Information Criterion) decides whether two or three states best fit. Across all 24 configs, BIC chose three states: favorable, neutral, unfavorable.

Walk-Forward Validation

We validate with walk-forward testing. Train the HMM on the first 70% of trades (about 875 out of 1,250), predict on the last 30% (about 375 trades). If the favorable state in the test data still produces better outcomes than the unfavorable state, the model generalizes. If not, it's overfitting.

Eleven out of 24 configs pass this test. This is the final filter before any trade management. If the HMM says unfavorable, you don't take the trade. Doesn't matter what the stop or the target would be.

Write this down. The HMM filter is the last gate before you risk capital.

The Complete Picture

Here's what we know before we've set a single parameter.

From step four: the full MAE and MFE distribution at P10 through P95, across 1,250 five-year measurements, 250 one-year measurements, and 60 ninety-day measurements, at one-hour, two-hour, and three-hour holds. From step five: this is a momentum edge, rho = 0.85. From step six: the edge flips in volatile markets, rho climbs to 1.20 across 750 volatile-regime trades. From step seven: today's conditions are favorable, validated on 375 out-of-sample trades.

Now we can talk about stop loss. It's not a feeling. In the current regime, for this edge type, at the one-year calibration window (250 trades): MAE P90 is 0.22%. That's where 90% of all trades stopped moving against us. That's the stop.

Take profit: MFE P60 is 0.18% across 250 one-year trades. Sixty percent of winners reach this level. Cover the queen: MFE P30 is 0.08% across the same 250 trades. Seventy percent of trades get there; take partial, let the rest ride. Hold time: rho stays below 1.0 through three hours in calm markets. Hold the full session.

Every number is measured. Every number has a statistical foundation. Every number adapts to market conditions.

Recap

Fixed constants give you consistent measurements. MAE and MFE show you what price actually does. The Markov chain tells you what kind of edge you have. The regime model tells you when that edge changes. That's what happens before I consider a stop loss.

If you want to see how we validate all of this, the Monte Carlo simulation, the walk-forward testing, the perturbation analysis that checks if your edge holds when you shift the stop by 25% across all 1,250 measurements, that's the next article. The measurement phase gives you the data. The validation phase tells you if you can trust it. Enlist in the bootcamp at thedailyprofiler.com to get the full 8-step pipeline.