Trading System Optimization

Initial Review

At this time, we examine the strategy for face validity, and ensure that the model doesn’t suffer from any biases that are standard in finance or portfolio theory, such as data snooping bias or survivorship bias. From there, we briefly discuss the means of trade execution in your system, which could be manual trade execution, automated trading, or a combination of both. We also discuss whether absolute return or risk-adjusted return is the objective of the trading model, its historical performance (if any) thus far, and the availability of the needed data. If we find that your system is in the very initial phases of construction, our team undertakes the more complete process described in the Trading System Development portion of our site.

Once we feel comfortable that we understand the core goal of your trading model, and before we begin working on the formal quantitative analysis, we outline for you qualitatively what we believe to be the most effective way to conduct the optimization. We also provide you with a brief report on the quantitative methodologies that we will use, based on the particular needs of your model.

Data Considerations

Our clients come to us with trading systems in various forms. Some clients approach us with a working model that they are already trading. These models are often specified in a trading software package such as Tradestation, Metastock, or E-Signal. Other clients come to us with their model specified in a statistical programming language such as R or SAS, or in a standard programming language such as C++. However, we also have many clients who come to us with trading systems which are simple to code and which have more trivial trade execution needs. In these cases, we can work with models relayed to us in something as straight-forward as a Word document with the various indicators and decision nodes indicated in text.

As most trading system optimizations involve robust backtesting and sensitivity analyses, having a reliable and cleaned data set is of extreme importance. Many of our clients use external data feeds which provide real-time data for trading, but do not have historical databases included. In these circumstances, we often have the data needed to do more extensive backtesting, and/or can acquire it at a reasonable cost. If the data is in a difficult to use format, we have a data cleaning team who can convert the data set into something more useful.

Backtesting and Optimizing the Trading Model

When it’s time to do the needed quantitative analyses, our primary objective is that all of our work be both robust and fully replicable. We don’t help you to make trading profits by creating overspecified versions of your model, by testing dozens of indicators, or by testing multiple variations of the same core model until we find something that “looks good”.

Our consulting team has wide experience in testing and optimizing trading systems, both academically and on Wall Street. They often confer over academic work that they are publishing, about the cutting edge strategies being developed every day in academia, or over their own models.Every member of our team knows that strategies need to be clearly specified beforehand in order to be robust, that multiple optimization attempts cannot be attempted on the same data set, and that violation of principles such as these will render valueless any predictions on the future performance of the model. This problem (data snooping resulting in non-robust results) is the issue that comes up the most for us when analyzing a pre-existing trading model. Many of our clients come to us not fully understanding the biases that can be present in financial market data.

We consider ourselves to be teachers and not just doers, and we will remain with you via phone, email, or in-person until you are completely comfortable with your understanding of these issues. If your model is failing due to them, we can attempt to respecify the model or find a way of creating a similar model that does not have these same pitfalls. The worst thing that can happen to a trading model developer is for he or she to inaccurately conclude that they have a profitable model, and to then lose both time and money trading that system until abandoning it for lack of success. Trading models that are created properly work in real-time, and not just in the past.

We use every tool in the portfolio theorists toolbox in order to optimally analyze (and revise if needed) your trading model. Due to our Wall Street and academic experience (every member of our team was at one point a professor of statistics, econometrics, economic theory, or finance), we are knowledgeable in almost every quantitative technique that is commonly used (and not so commonly used) in trading model validation and optimization, including but not limited to:

  • Time Series Analysis
  • Kalman filtering
  • Principal Components Analysis
  • Monte Carlo Simulation
  • Jackknife/Bootstrapping Techniques
  • General Autoregressive Conditional Heteroskedasticity
  • Black-Scholes Option Pricing
  • Simulation Based Option Pricing
  • Advanced Derivative Valuation
  • Path Dependent Security Pricing
  • Brownian Motion and Geometric Brownian Motion
  • Interest Rate Term Structure Modeling
  • Spot Rate and Overall Volatility Term Structure Modeling
  • Single-Factor Markovian Short-Rate Modeling
  • Time Varying Volatility Analysis
  • Securitization Analysis, Pricing, and Valuation (ABS, MBS, CDS)
  • CDS Pricing and Valuation
  • Steepest Ascent Optimization
  • Ito calculus
  • Simulated Annealing
  • Exhaustive (“Brute-Force”) Optimization
  • Covariance Matrix Adaptation-Evolution Strategy
  • Genetic Programming
  • Machine Learning Algorithms (Gradient Boosted Models, Random Forests, Support Vector Machines, etc.)

Deliverables and After-Support

When our core work on your trading model optimization is complete, we ordinarily send a final deliverable report which includes the entirety of our analyses, results, syntax code to replicate the analyses, and of course a full recommendation. If the model could benefit from optimization, we would of course provide you with the optimized model in any format of your choice. We will also continue to work with you on that model as you begin trading it again, so that we can make any adjustments needed going forward, use more data from the new trading to confirm that our assumptions have proven accurate, run a power analysis in continuous time to ensure that we have sufficient sample size to validate the second run of your model, and suggest other models that may work well in concert with your current model(s). We also would provide you with a risk-management review and further implementation recommendations.

If we conclude that the model is unlikely to be profitable and all backtests and other follow-on analyses confirm this, then we would attempt to take the substance of your model and suggest other types of models that could be specified robustly and target that same idea. Of course, we remain available for consultation on any of these issues, either at that time or in the future. Trading models can’t be carried out by rote (this is true of even automated trading systems), they need to be understood conceptually. We will remain with you, in a professorial role that is still familiar to many of us, until you have that conceptual understanding.


We understand the importance and necessity of confidentiality when dealing with any trading model or idea, and provide all of our prospective clients with a Non-Disclosure Agreement immediately upon contact. This ensures you that your idea and/or existing model will not be shared with any third parties, and your consultation with us is completely confidential.

Please click here to read our Data Security Policy.