Most traders approach strategy development by deciding what they think should work and then testing whether the data agrees. Brian Miller does the opposite. He starts with the data, extracts the market’s characteristics, and builds a strategy around what the R&D reveals. Game theory is one of the philosophical frameworks guiding that process.
Brian is the founder of Optimized Trading, a quantitative firm focused on systematic, short-to-long-term swing strategies across index futures, ETFs, and other instruments. This is the second part of his BST conversation, expanding on model development, dynamic analysis, control modules, and why creativity may be the most underrated tool in a systematic trader’s toolkit.
Watch the full episode below, then read on for the complete breakdown.
What game theory actually offers systematic traders
Game theory is commonly associated with HFT and scalping, but Brian applies its philosophical principles at longer timeframes. The underlying idea is that markets reflect human behavior, and human behavior follows patterns that can be modelled.
His practical application: he looks for the behavioral signals of large institutional buy-and-hold investors. Companies like Walmart, Johnson & Johnson, and Procter & Gamble attract institutional capital when the market seeks protection, and that capital moves predictably. When institutions see growth opportunities, they shift toward growth stocks; when they perceive risk, they shift back into consumer staples and large-cap defensives.
By studying these rotation patterns systematically, he builds strategies that are timed to follow institutional flows rather than predict them. The larger the capital pool behind a behavior, the more reliably it produces a tradeable signal.
Never go into development with your mind made up
The most common error Brian observes in experienced strategy developers: they already have a bias toward what will work before they start. They’ve had success with momentum strategies in the past, so every new project gets unconsciously pushed toward a momentum framework, regardless of whether the new instrument’s data supports it.
His solution is procedural. At the start of every new project, he dives into the raw data of the instrument or asset class he’s working with. He examines the market’s natural cycles, frequency strengths, whether it has more reverting or momentum characteristics, and its correlation structure relative to other instruments he trades. Only then does he begin to form a concept for what might work.
He creates a development guideline at the start of each project specifically to document what the R&D has revealed before any strategic choices are made. This serves as a check against later bias, where it becomes tempting to adjust the framework when early results don’t match expectations.
Dynamic analysis: letting market cycles set the parameters
Static parameters are one of the most common sources of overfitting. A look-back period that worked well historically may be wrong for the current market environment. Brian’s approach is to let the current market data determine key parameters dynamically.
For a multi-frequency momentum strategy, instead of fixing the look-back periods at set values, he correlates them to the current cycle strengths in the market. If small, medium, and large frequency momentum are all at extreme readings, the strategy uses look-back values derived from what those conditions have historically looked like, not from a fixed optimisation.
He extends this to exit models. Rather than applying a single exit rule, he deploys a range of exit strategies, both profitable exits and loss exits. Dynamic analysis helps determine which exits are most appropriate for the current market condition.
Simplicity at the model level, complexity at the control level
Brian draws a specific distinction between where simplicity and complexity belong in a systematic structure. At the individual model level, he aims for what he calls “creative simplicity.” Simple strategies are more robust and carry fewer risks of failure. This aligns with John Conway’s work on how robust solutions tend to be simple ones.
But complexity has a place in the control structures that sit above individual strategies. A multi-model system can use machine learning, correlation analysis, and decision trees at the control level to decide which models are active, how they are weighted, and how the portfolio responds to changing market conditions. The key distinction is that at the control level, you still know what the underlying models are doing, so you can identify the risks of any given model even if the control mechanism is sophisticated.
Using machine learning to build the model itself carries a specific risk: you may not understand why it works, which means you can’t identify when conditions have changed enough that it should stop working.
Correlation analysis: the single most valuable control mechanism
When asked to choose one control mechanism above all others, Brian is direct: correlation analysis. It provides both reactive and predictive information. It tells you how closely a model’s behavior currently matches the market condition it was designed for, how models perform relative to each other across different environments, and how far price moves from entry before validating a position.
He builds two types of weighting in his control module: current market environment identification and projected market environment identification. The projected element is probabilistic, not certain, but weighting the control module toward the likely future environment, while still being able to respond quickly to unexpected changes, produces better outcomes than relying entirely on either predictive or reactive logic alone.
Price data evolution and why the post-1998 era requires a different approach
Brian does not use data before approximately 1998. A fundamental shift in market characteristics occurred around that time, driven by the growth of electronic trading and the increasing presence of algorithmic participants. Markets since then contain more randomness than before, and he doesn’t expect that to reverse.
His indicator for whether a strategy is handling this evolution well: look at the year-over-year performance of the strategy across its full history. A healthy strategy shows steadily improving correlation to the market over time as the market’s characteristics move closer to what the strategy was designed to exploit. A declining trend in year-over-year performance, even if the overall equity curve still rises, is a warning that the strategy is slowly becoming less aligned with market structure.
His response to this challenge is structural rather than strategic: don’t be dependent on any single model, and don’t rely on a single entry point for any position. Allowing multiple entries to exploit a condition diversifies the factors that produce outcome and reduces the damage when price data evolution starts eroding one approach.
Creativity as a trading edge
Brian’s closing point is the one that surprised me most. He argues that traders who follow conventional rules about what makes a good strategy, such as requiring a high profit factor, a Sharpe above a certain threshold, or mandatory stop-loss levels, are constraining their own search space. Some of his best strategies have violated these rules.
He recommends deliberately detaching from the received wisdom of strategy development and letting the R&D process go where it goes. The market is a combination of human behavior, corporate dynamics, market cycles, and structural factors. The number of places from which alpha can be extracted is large. You are only limited by your own level of creativity, and traders who accept inherited constraints have already limited the search.
Related episodes
- 3 Under-Utilized Techniques to Build Better Trading Strategies – Brian Miller
- Laws of Trading: Better Decision-Making for Traders – Agustin Lebron
- Market Behavior – Adam Grimes (Episode 55)
- Finding an Unfair Advantage – Tomas Nesnidal (Episode 52)
Get the show notes & transcript
Want to learn more about building adaptive systematic trading strategies? Subscribe to the Better System Trader podcast for weekly interviews with the world’s top systematic traders and quantitative researchers.
