Compare Evolver and Microsoft Excel's Solver

Solver works on a gradient descent principle, whereby it finds optimal locations on the cost function surface measuring local gradients (slopes) on that surface and moving toward the best spot. Movement can be constrained by additional requirements, such as specifying that x+y>2. The cost surface must be continuous (smooth) for this process to work.

Evolver works on a completely different principle. It does not care what the surface looks like. It could be discontinuous, filled with "potholes", whatever. Evolver optimizes by comparing the performance of numerous potential solutions, throwing away the bad ones, and mutating the rest. This creates a new set of potential solutions and the process is repeated, over and over, with the solutions getting a little better each time.

As for building a financial trading system, slight changes in a system parameter could conceivably create drastic changes in system performance. We'd rather not have a system with that discontinuous property, but when we are starting from nothing and expect Evolver to create something, such discontinuities will be prevalent. Consequently, Solver would not be as successful for building financial trading systems.

What data do I need to supply Evolver, in order to produce a good trading strategy?

Imagine you have a trading system either on your spreadsheet or written in Excel's visual basic. The system includes rules and each one is governed by one or more parameters, such as scaling factors, threshold values, etc. Your trading system can then be represented by the entire collection of all parameters that modify these rules. Evolver can be set up to modify these parameters within specified ranges, in order to achieve optimal trading performance. You can define optimality any way you like, including net profit, return on account, profit factor, etc.

How many data records do I need to supply ?

That depends on the complexity of the process you are trying to model, ... a trading system, for example. Simple systems require less training data than complex systems, simply because the latter has more parameters to adjust.

If your goal is to produce a trading system, then you must consider how well this system will continue to perform into the future. The more statistically valid your estimate is now, the more you can expect similar performance later. To attain a statistically valid estimate of system performance, you need enough data to produce at least 100 trades, or 20 times the number of parameters you are using Evolver to optimize, whichever is greater.

For example, suppose your trading system has 13 parameters to be optimized. Then the final system will need to produce at least 260 trades (20 x 13) on the historical data for any measure of its performance to have statiscial validity.

Does Evolver come with a guarantee ?

Yes. The developer offers 30 day money-back guarantee. NOTE - Jurik Research will charge a $25 "restocking" fee.

Is there a software review on Evolver ?

PC World says Evolver "Is an excellent choice for complex profit-optimizing" and "impressively fast on big problems". 

Charles Seiter, author ---

Microsoft Excel can’t do everything. If you’re dealing with what-if situations in production scheduling, workflow, logistics, and financial modeling, you need more sophisticated tools. Excel’s built-in Solver can handle some assignments, such as trying to optimize profits across a product line, but it’s generally limited to situations with smoothly varying quantities- --expects you to accept answers like "3.7 trucks" in shipping problems.

On the other hand, Evolver, an Excel add-on from Palisade, is well suited to handling optimization in many real-world problems where the basic information involves many variables, as in shipping and manufacturing. I tested a shipping version.

Evolver implements a genetic algorithm. The optimizer generates a batch of random inputs as variables for the problem. Then it checks the results that the variables produce, keeps the variables that yield the best results, and creates a new generation of inputs by mixing components of surviving variables. Also, Evolver works impressively fast on big problems (with up to 80 variables). Check this program out if you’ve outgrown Solver

PRO: Fast, sophisticated analysis tool for use with Excel.
CON: None significant
VALUE: Excellent choice for complex profit-optimizing