Institutional Strategy Development Framework

Built For

Instruments: Multi assets
Trading Style: Day Trading / Swing Trading

Strategy Overview

The main idea behind this framework is simple: institutional traders do not start by looking for the next trade. They start by looking for a repeatable strategy that can be tested and validated with data. A retail trader may find a setup, see that it worked a few times, and move straight into live trading.

The institutional process adds several steps before real money is involved.

Idea → Rules → Code → Backtest → Validation → Live Trading → Monitoring

The goal is not to prove that the next trade will win. The goal is to understand whether the strategy has a positive expected return when the same rules are repeated over a large number of trades.

A strategy can have a 70% win rate and still lose on the next trade. The edge only becomes meaningful across a larger sample.

In the framework, Matteo uses four market ideas to show how this process works:

• Opening Range Breakout
• VWAP Trend Strategy
• Post-Earnings Announcement Drift
• Overnight Market Anomaly

The bigger lesson is not to copy these strategies exactly. It is to understand how an institutional trader takes an idea and turns it into something that can actually be tested.

The Institutional Trading Process

Start With a Trading Idea

Every strategy begins with a hypothesis about how the market behaves. Instead of randomly searching charts for patterns, Matteo explains that institutional traders often start with financial research.

One source discussed is the Social Science Research Network, where academic researchers and market professionals publish studies around topics such as momentum, mean reversion, volatility, earnings, and market anomalies.

The important rule is:

Do not copy the research paper directly.

Use the research to understand:

• What was discovered?
• Why might that behavior exist?
• Can that idea be turned into clear trading rules?

The research provides the foundation. The trader still has to build and test the strategy.

Academic vs Industry Research

Matteo separates research into two broad categories.

Academic research is generally produced by universities, professors, and researchers.

Industry research can come from hedge funds, traders, portfolio managers, and other market professionals.

Both can provide useful strategy ideas. The goal is not to find a paper and assume it will continue working forever. The goal is to extract the underlying market behavior and test whether that behavior can still be turned into an edge today.

Define the Strategy Rules

Before anything can be tested, the strategy needs to be completely objective. Matteo explains that every strategy needs three components.

Entry

The entry defines exactly why the position should be opened. The condition should be clear enough that there is no interpretation involved.

For example:

Go long when a one-minute candle closes above VWAP.

That is much more objective than saying:

Buy when price looks strong around VWAP.

Exit

The strategy also needs a clear reason for closing the trade.

An exit can include:

• Stop loss
• Take profit
• Time-based exit
• Price or indicator condition

A time exit is especially important in some intraday systems. For example, if neither the target nor stop has been reached by 3:30 PM, the strategy may automatically close the position.

Position Sizing

The final rule is how much capital should be placed into the trade. Matteo emphasizes that position sizing should be part of the strategy itself. It affects both risk and overall performance.

During early development, he suggests keeping this simple, usually by testing one fixed contract. More advanced sizing can be added once the underlying strategy has shown that it has an edge.

Encode the Strategy

Once the rules are clear, the next step is translating them into code. The purpose of coding is not necessarily to become a fully automated trader. It allows the strategy to be tested consistently across historical data.

Matteo explains that large language models have made this much more accessible to retail traders. The recommended approach is to build the strategy gradually.

Build One Rule at a Time

Start with the entry condition.

Check that the code is entering exactly where expected.

Then add the exit.

Test it again.

Then add position sizing.

Continue until the coded version behaves exactly like the original rules.

The important point is that the trader still needs to understand what the code is doing.

AI is a tool that helps with implementation. It does not replace the trader’s understanding of the strategy.

Backtest the Strategy

Backtesting means applying the exact strategy rules to historical market data. Every time the historical market satisfies the entry conditions, the system records a simulated trade. This allows years of trading history to be tested without manually scrolling through charts.

What the Backtest Should Show

The backtest gives the trader an equity curve and statistics such as:

• Net profit
• Win rate
• Average trade
• Average winning trade
• Average losing trade
• Maximum drawdown
• Losing streaks

These statistics show whether the strategy is moving in the right direction. However, a profitable backtest alone is not enough.

A strategy can look extremely good historically simply because the rules were fitted too closely to that specific historical period. That is why the next stage is validation.

Validate the Strategy

Validation is used to determine whether the strategy has a genuine edge or whether the historical performance came from overfitting or luck. Matteo explains two main forms of validation.

In-Sample vs Out-of-Sample Testing

Historical data is divided into two sections.

A simple example discussed in the episode is:

80% In-Sample / 20% Out-of-Sample

The in-sample section is where the strategy is developed.

This is where the trader can:

• Add rules
• Remove rules
• Adjust parameters
• Test different variations

Once the strategy has been developed, the rules are frozen. The exact same strategy is then tested on the remaining unseen data. That is the out-of-sample test.

If the strategy performs well during development but collapses on unseen data, it may have been overfit.

What Overfitting Means

Overfitting happens when too many rules or parameters are added specifically to make historical results look better. The strategy becomes very good at explaining the past but may have little ability to handle future market conditions. A stronger strategy should remain reasonably stable when applied to data that was never used during development.

Monte Carlo Analysis

After the in-sample and out-of-sample test, Matteo adds another validation layer through Monte Carlo analysis.

The goal is to understand how much the strategy’s results depend on the exact sequence of historical trades.

Monte Carlo Shuffle

The simplest version keeps every historical trade but changes the order in which those trades happened.

For example, the original backtest may show a maximum losing streak of 15 trades.

Once the same trades are randomly rearranged thousands of times, some simulations may produce larger losing streaks.

The strategy itself has not changed.

Only the order of the trades has changed.

If most simulated equity curves remain close together, the strategy may have a more stable underlying edge.

If the outcomes spread dramatically apart, the strategy may be more fragile.

Monte Carlo Resampling

The second version goes further.

Some historical trades can appear more than once in a simulation while others may not appear at all. This produces thousands of different possible equity curves.

The output can help estimate:

• Expected return
• Expected profit
• Potential drawdown
• Maximum losing streak
• Probability of experiencing a specific loss

These numbers can later be used as reference points during live trading.

Using Monte Carlo During Live Trading

Suppose the simulation shows that there is only a very small probability of experiencing more than a $10,000 drawdown after ten trades. If the live strategy suddenly exceeds that level, it can act as a warning.It does not automatically mean the strategy is broken. But it tells the trader that the live performance is starting to move outside the range that historical testing suggested was normal.

Strategy Model 1: Opening Range Breakout

The first practical example is built from research around Market Intraday Momentum.

The Research Finding

The research looks at the overnight move together with the first 30 minutes of regular trading. The idea is that this period can reveal the imbalance between buyers and sellers.

New information accumulates while the market is closed.

That can include:

• News
• Earnings
• Global market movement
• Overnight positioning

Once the regular market opens, many of those orders begin entering at the same time.

The first 30 minutes can therefore reveal which side has the stronger imbalance.

If that imbalance is large enough, it may continue influencing price later in the day.

Define the Opening Range

The range is created between:

9:30 AM and 10:00 AM Eastern Time

Mark:

Opening Range High

and

Opening Range Low

Matteo explains that the 30-minute period is the research-backed starting point for this idea.

Long Entry

Once the range has been created:

Go long when price closes above the opening-range high.

The rule uses a close above the range rather than simply a touch.

Stop Loss

Place the stop at:

The opening-range low.

This gives the trade a clearly defined invalidation point.

Take Profit

Matteo suggests testing simple reward-to-risk targets.

The starting range is:

1:1 to 1:2 reward-to-risk

The strategy is deliberately kept simple during development so the trader can clearly understand which part of the model is creating the edge.

Time Exit

If price has not reached the stop or target:

Close the position around 3:30 PM ET.

This keeps the strategy intraday.

Short Version

The opposite version can also be tested:

Short when price closes below the opening-range low.

However, Matteo explains that from his experience, this particular imbalance concept has historically performed better on the long side.

Buyers can step into downside breaks and remove the imbalance more quickly.

Because of that, the short strategy should be tested separately instead of assuming it will perform like the long version.

Opening Range Position Sizing

During development:

Use one contract.

This keeps the backtest simple.

But one fixed contract can create very different levels of risk depending on the day’s volatility.

A wide opening range could create a much larger dollar loss than a narrow opening range.

For live trading, Matteo prefers the idea of volatility-adjusted sizing.

Volatility-Adjusted Risk

The goal is to keep the amount of money being risked relatively consistent.

High volatility:

Use fewer contracts.

Low volatility:

Use more contracts.

This prevents a small number of highly volatile sessions from controlling the entire strategy’s performance.

Opening Range Breakout Rules

Long Setup

Range: 9:30-10:00 AM ET
Entry: Close above opening-range high
Stop: Opening-range low
Target: Test 1R-2R
Time Exit: Around 3:30 PM ET
Development Size: 1 contract
Live Size: Consider volatility-adjusted sizing

Short Setup

Entry: Close below opening-range low
Stop: Opposite side of the range
Management: Test separately from the long model

The episode specifically warns that the short side may not contain the same edge.

Strategy Model 2: VWAP Trend Strategy

The second example uses research around VWAP.

VWAP stands for Volume Weighted Average Price.

It represents the average price traded during the session, weighted by volume.

The Research Finding

The basic model studied was very simple.

Using a one-minute timeframe:

Go long when price closes above VWAP.

Go short when price closes below VWAP.

Then exit when price crosses back through VWAP.

The research discussed used QQQ, but Matteo’s focus is on understanding the logic and then testing that idea across other markets.

Why VWAP Matters

VWAP is widely used in institutional execution.

Large market participants often use execution algorithms designed to participate with market volume while achieving an average execution close to VWAP.

When trading activity becomes heavy, these algorithms can send more orders into the market.

Because many institutional participants use similar execution benchmarks, activity around VWAP can contribute to directional movement.

The edge therefore has a possible market-structure explanation instead of VWAP simply being treated as another indicator.

Long Entry

Using the one-minute research model:

Go long when a one-minute candle closes above VWAP.

Short Entry

Go short when a one-minute candle closes below VWAP.

Exit

For either direction:

Close when price crosses back through VWAP.

The core version does not require a separate take-profit target.

The VWAP recross itself provides the exit.

VWAP Session

Matteo’s preferred starting point is to calculate VWAP from the regular market open.

For US markets:

Start VWAP at 9:30 AM ET.

Other versions, such as using previous-day VWAP, can be tested later. They should not automatically be added without validation.

Markets to Test

Although the original research example used QQQ, Matteo explains that the concept can be tested across markets where institutional VWAP execution is relevant.

Examples include:

• NQ
• ES
• Crude oil
• Individual stocks

The important point is to confirm through testing that the edge exists on the instrument being traded.

VWAP Strategy Rules

Long

Timeframe: 1-minute starting point
VWAP: Session VWAP from regular market open
Entry: Candle closes above VWAP
Exit: Price crosses back through VWAP

Short

Entry: Candle closes below VWAP
Exit: Price crosses back through VWAP

Position Sizing

During development:

Use one contract.

Once the underlying edge has been confirmed, volatility-adjusted sizing or portfolio-based sizing can be tested.

Strategy Model 3: Post-Earnings Announcement Drift

The third model is based on the Post-Earnings Announcement Drift, often shortened to PEAD.

This is one of the longest-studied market anomalies discussed in the episode.

The Core Idea

When a company releases earnings that significantly surprise the market, the stock does not always fully adjust immediately. Instead, price can continue drifting in the direction of the surprise over the following days or weeks.

Positive earnings surprise + positive price response:

Potential upward drift

Negative earnings surprise + negative price response:

Potential downward drift

Earnings Surprise

The model does not rely only on price.

It also compares the company’s actual earnings per share with what analysts expected.

Positive surprise:

Actual EPS > Expected EPS

Negative surprise:

Actual EPS < Expected EPS

Matteo explains that combining the earnings surprise with the actual market reaction gives the strategy a stronger foundation.

Why the Drift Exists

Two main explanations are discussed.

Slow Information Flow

Mega-cap stocks receive enormous analyst and investor attention.

New information can therefore be reflected in price very quickly.

Small-cap and mid-cap stocks often receive less coverage.

Information can take longer to spread through the market.

That creates the possibility of a slower price adjustment.

Liquidity Constraints

Large funds may want to build a significant position after strong earnings. However, they may not be able to buy the entire position immediately without moving the stock too much. Instead, purchases can be spread across several days or weeks. That continued institutional demand can help create the drift.

PEAD Long Setup

The simple research starting point is:

Positive price reaction after earnings

AND

Actual EPS is above expected EPS

Then:

Enter long at the opening of the following trading day.

Exit

The research discussed suggests using approximately:

60 trading days

as the baseline time exit.

Matteo recommends beginning with this simple version before adding technical stops or additional conditions.

Adding a Stop

A stop loss can be researched later.

Examples discussed include:

• Percentage-based stop
• Pre-earnings closing price
• Other risk-management levels

These are examples to test rather than fixed rules from the original model.

PEAD Position Sizing

Several starting approaches are discussed.

A trader could allocate:

1-2% of the portfolio to each stock

or use a fixed dollar allocation.

More advanced models could change the position size based on:

• Size of earnings surprise
• Size of price reaction
• Volatility

Each version needs to be tested separately.

Where PEAD May Work Better

The effect may be weaker in heavily followed mega-cap companies because information is processed quickly.

Matteo highlights areas such as:

Small-cap stocks

Mid-cap stocks

and potentially:

European equities

as areas where the idea may be worth researching.

The logic is that lower analyst coverage and lower liquidity can cause information to travel more slowly.

Short-Side Limitation

The short side may be less effective today.

Companies sometimes warn investors about poor performance before official earnings are released.

If the market already expects weak numbers, the negative surprise becomes smaller.

Part of the move may already have happened before earnings.

That means the downside drift can be weaker.

For that reason, the long and short versions should be evaluated independently.

PEAD Strategy Rules

Long Setup

Universe: Small/mid-cap stocks as primary research candidates
Catalyst: Earnings announcement
Condition 1: Positive price reaction
Condition 2: Actual EPS > Expected EPS
Entry: Following trading day’s open
Baseline Exit: 60 trading days
Position Size: Test portfolio percentage, fixed dollars, or volatility/surprise-adjusted sizing

Additional filters should only be added if testing shows they improve the strategy.

Strategy Model 4: Overnight Market Anomaly

The final research example examines where long-term stock-index returns actually occur.

The Research Finding

The comparison is between:

Regular Trading Hours: 9:30 AM → 4:00 PM

and

Overnight: 4:00 PM → 9:30 AM

Research discussed in the episode found that a surprisingly large portion of historical equity-index returns occurred during the overnight period rather than during regular trading hours.

Matteo explains that similar behavior can also be observed when testing more recent Nasdaq futures data.

The important lesson is not that a fixed percentage will always remain the same.

The finding provides another market behavior that can be turned into a strategy hypothesis.

Basic Overnight Test

The simplest test is:

Enter long at 4:00 PM ET

and

Exit at 9:30 AM ET the following day.

The purpose of this model is initially to measure whether the overnight effect actually exists.

It is not presented as a fully finished live strategy.

Why the Overnight Effect May Exist

Unlike the previous examples, Matteo explains that there is no single agreed explanation.

Two main theories are discussed.

Overnight Risk Premium

Holding a position overnight comes with additional uncertainty.

Liquidity is lower and new information can arrive while regular markets are closed.

Investors may therefore receive additional return for accepting that risk.

Lower Overnight Liquidity

News and earnings frequently arrive outside regular trading hours.

Because fewer participants are trading, the same amount of buying or selling can create a larger price move.

Once regular liquidity returns, the market may partially reverse toward fair value.

Building Strategies From the Overnight Effect

The important part of this example is the research process.

If the trader discovers that much of the directional movement happens overnight, the next question becomes:

Can a strategy be specifically designed to capture that movement?

For example, a trader could research an overnight breakout or momentum model rather than using the same opening-range concept only during regular market hours.

The process becomes:

Research Finding → New Idea → Rules → Code → Backtest → Validation

Overnight Starting Rules

Baseline Model

Entry: Long at 4:00 PM ET
Exit: 9:30 AM ET the following trading day

Matteo does not provide a finalized stop-loss or profit-target model for this strategy.

Those rules would need to be developed and validated separately.

Position Sizing and Volatility Targeting

Position sizing is one of the most important concepts throughout the episode.

Using the same number of contracts on every trade does not necessarily mean the same amount of risk.

A highly volatile market may create a very wide stop.

A quiet market may create a very small stop.

Trading the same position size in both situations can make the volatile trade dominate the strategy’s performance.

Volatility Targeting

The alternative is to begin with the amount of money the strategy is willing to risk.

Then calculate position size from the current volatility.

For example:

Target Risk = $10,000

High volatility:

Smaller position

Low volatility:

Larger position

The goal is to keep the strategy’s risk more consistent from trade to trade.

During early development, Matteo still recommends starting with one contract so the trader can first understand whether the underlying strategy itself works.

Keep the Strategy Simple

One of the strongest points from the episode is that complexity does not automatically create better performance.

More rules can make a strategy more fragile.

Every additional condition creates another place where the strategy can fail or become overfit.

The preferred process is:

Start simple.

Prove that the basic edge exists.

Then test whether additional rules actually improve the results.

A simple strategy with a clear market reason behind it can be more robust than a complicated system made from dozens of indicators.

Understand Why the Edge Exists

Before optimizing a strategy, understand the market behavior it is trying to capture.

For the Opening Range Breakout:

Early-session order imbalance

For VWAP: Institutional execution around volume-weighted price.

For PEAD: Slow incorporation of unexpected earnings information.

For the Overnight Anomaly: Returns occurring disproportionately outside regular trading hours.

Knowing the reason behind the strategy also helps the trader recognize when the market structure supporting that strategy may have changed.

Using AI in Strategy Development

AI can help reduce the technical barrier between finding an idea and testing it.

It can be used to:

• Understand research papers
• Extract the main findings
• Explain the possible market logic
• Translate rules into code
• Create strategy variations

However, Matteo treats AI as a tool rather than something that should be trusted blindly. The trader still needs to verify every part of the strategy.

Build Iteratively

The recommended sequence is:

Entry first.

Test it.

Add the exit.

Test again.

Add position sizing.

Test again.

Continue until the complete strategy behaves exactly according to the intended rules.

Automation Is Not Required

Encoding a strategy does not mean the trader has to automate everything.

There are several ways to use the process.

Research Only

Code the strategy so it can be properly backtested, then continue trading manually.

Confirmation

Use the systematic model as another layer of confirmation for a discretionary setup.

Full Automation

Allow the computer to monitor the market, enter, manage, and close positions according to the rules.

The important advantage is the ability to objectively measure whether an idea has evidence behind it.

Build a Portfolio of Strategies

The institutional end goal is not necessarily to find one perfect strategy. It is to build multiple strategies that behave differently.

A momentum strategy may perform very well in a trending market but struggle during sideways conditions.

A mean-reversion strategy may perform better during those same sideways periods.

Combining strategies with different return drivers can create more consistent overall performance.

How Many Strategies?

For a retail trader, Matteo suggests working toward roughly: 3-4 strategies running at the same time. The purpose is not simply to trade more.

The strategies should be uncorrelated, meaning they should not all depend on exactly the same market environment. At the institutional level, automation can allow hundreds of strategy variations to run simultaneously across different markets.

Monitor the Strategy After Deployment

A strategy being profitable historically does not mean it will work forever.

Market structure changes.

Liquidity changes.

Participant behavior changes.

The original inefficiency may become weaker or disappear.

That means the process does not end when the strategy goes live.

The trader should continue monitoring: Performance and risk

If live results begin falling well outside the range predicted by historical testing and Monte Carlo analysis, the strategy may need to be reviewed.

Complete Strategy Development Process

Step 1: Find the Idea

Use research to identify a repeatable market behavior.

Step 2: Understand the Why

Determine why that behavior may continue to exist.

Step 3: Define the Rules

Every strategy needs:

Entry + Exit + Position Sizing

Step 4: Encode

Translate the strategy into objective code.

Step 5: Backtest

Apply the rules to historical data.

Step 6: Develop In-Sample

Adjust and improve the strategy using the development dataset.

Step 7: Freeze the Rules

Stop making changes before looking at the unseen data.

Step 8: Test Out-of-Sample

Check whether the same strategy still performs on unseen history.

Step 9: Run Monte Carlo Analysis

Test different possible sequences and distributions of trades.

Step 10: Finalize Position Sizing

Determine how the strategy will control risk during different volatility environments.

Step 11: Deploy

Trade manually, semi-automatically, or fully automatically.

Step 12: Monitor

Compare live performance with what the strategy’s historical testing suggested should be normal.

Step 13: Build More Strategies

Develop additional uncorrelated strategies and repeat the process.