Parameterless Trading Models

A portfolio managing director that I used to piece of job for similar to pronounce that his trading models convey "no complimentary parameters". As is customary inwards our secretive industry, he would non elaborate farther on his technique.

Lately, I get to empathise what a trading model amongst no complimentary parameter means. It doesn't hateful that it does non incorporate whatever lookback menses for calculating trends, or thresholds for entry or exit. I recall that would hold upward impossible. It merely agency that all such parameters are dynamically optimized inwards a moving lookback window. This way, if you lot ask: "Does the model convey a fixed net turn a profit cap?", the trader tin honestly reply: "No, net turn a profit cap is non an input parameter. It is determined yesteryear the model itself."

The wages of a parameterless trading model is that it minimizes the danger of overfitting the model to multiple input parameters. (The so-called "data-snooping bias".) So the backtest functioning should hold upward much closer to the actual frontwards performance.

Now, it is quite computationally challenging to optimize all these parameters just-in-time for your side yesteryear side order, but it is oft fifty-fifty to a greater extent than hard to produce that inwards a backtest, given that a multidimensional optimization demand to hold upward performed for each historical bar. As a result, I personally convey seldom traded parameterless models, until I buy the farm to interrogation my regime-switching model. That model is nearly parameterless (I left out a few parameters from optimization because of a lack of time, non because of whatever technical difficulties).

The argue backtest optimization tin at in 1 lawsuit hold upward done inside a few minutes is due to my utilization of Alphacet Discovery's server-based optimization engine. There may hold upward other optimization software out at that spot that performs similar functions efficiently -- I welcome comments from the reader.

A Method For Optimizing Parameters

Most trading systems own got a issue of parameters embedded, parameters such equally the lookback period, the entry in addition to perish thresholds, in addition to then on. Readers of my weblog (for e.g., here in addition to here) in addition to my book would know my persuasion on parameter optimization: I am no big fan of it. This is because I believe fiscal fourth dimension serial is likewise non-stationary to permit i to enjoin what was optimal inwards the backtest is necessarily optimal inwards the future. Most traders I know would rather merchandise a strategy that is insensitive to modest changes inwards parameters, or alternatively, a "parameterless" strategy that is effectively an average of models amongst dissimilar parameters.

That existence said, if you lot tin solely merchandise i model amongst i specific fix of parameters, it is rational to inquire how i tin choice the best (optimal) fix of parameters. Many trading models own got a adept issue of parameters, in addition to it is quite onerous to detect the optimal values of all these parameters simultaneously. Recently, Ron Schoenberg published an article inwards the Futures Magazine that details a agency to achieve this amongst simply a tiny sum of figurer power.

The fundamental technique that Ron uses is cubic polynomial tally of the P&L surface equally a constituent of the parameters. Ron uses the VIX RSI strategy inwards Larry Connors' majority "Short Term Trading Strategies That Work" equally an example. This strategy has five parameters to hold upwards optimized, but Ron solely needs to compute the P&L for 62 dissimilar sets of parameters, in addition to the whole physical care for solely takes 58 seconds.

Although Ron has confirmed that nigh of the parameters that Connors picked are about optimal, he did detect a few surprises: namely, that RSI of menses three or iv is significantly to a greater extent than profitable than the ii that Connors used, at to the lowest degree inwards the backtest period.

Now, for a truthful essay out of this optimization, it would hold upwards helpful if Ron performed this optimization withholding some out-of-sample data, in addition to run into if these parameters are notwithstanding optimal inwards that withheld information set. Since he didn't produce that, nosotros demand to hold off for some other yr to detect out ourselves!

How Is The Chore Marketplace Rate For Quants These Days?

Felix Salmon claimed inward parameterless trading models similar I advocated, 2008 is non that odd except for its higher volatility.