The growth of automated trading and development has led to a heightened degree of awareness towards the pitfalls of over optimisation. Curve Fitting as it is known, has long been a problem caused by over optimisation of parameters and technical indicators to provide a favourable outcome. In essence, an online forex trader has to be careful when defining a strategy based on a specific period or periods. A curve fitting scenario can lead to significant losses and will in most cases differ completely on a performance basis.
The difficulty of curve fitting strategies is the definition process and the ability to weed out the profitable from the non profitable. Some important ways to identify whether a strategy is curve fitted includes:
- The testing of the strategy by utilising random periods or data. If the performance is completely different or does not correspond to overlayed periods, then the strategy could be curve fitted.
- Over optimisation can result in discrepancies in performance and validity. Traders that tweak indicators, and strategies, to the point where a profitable result is attained. An effective way of identifying whether this occurs is through the detailed analysis of the base indicators. Once again the performance data will provide an objective overlay and highlight the degree of over optimisation
To truly understand the concept of curve fitting, we have highlighted below two different examples, based on the moving average indicator.
Example 1:
Trader A is developing a system based on the basic moving average parameter. He chooses an entry logic when the short term (15) day moving average crosses over and below the long term (50) moving average. The Trader wishes to add a stochastic overlay as an exit point, with parameters which include (15,2,50). These define the level of range and how sensitive the indicator is over the specified time frame.
Trader A leaves the basic parameters and sees that the strategy is not profitable. He then generates a scan over the last 4 months, to determine the specific parameters that have been profitable. The Stochastics measurements highlight that (10,2,70)are the most profitable over this period, and the chart indicates a significant profit. This can be identified as curve fitted, as the parameters are determined by specific periods. In essence over the long run, these settings will not work and could provide for significant losses.
Example 2:
Trader B also wishes to develop a strategy based on the basic moving average indicator. Instead of using the stochastics overlay, she decides to focus on the Bollinger bands as entry and exit points. Once the indicator is overlayed with default settings, the profitability of the strategy is quite high. To determine the effectiveness of long term performance, Trader B modifies basic settings. Specifically choosing random data points, the Trader is then able to hypothesis. In this example the performance is consistent and deemed to be not curve fitted.
For more information, please visit Vantage FX.
