Off your game?

Read our free Dr K report on how to optimize your mind and body so you can boost your focus when trading the markets.
Your email will always remain private.

Frequently Asked Questions

Dr K VIX Volatility Model
Please explain how you did your backtests.

Q: I would appreciate knowing how many data points the model was tested against. As you know the number data points are more important than the number of years- if the number of signal changes per year is merely in the dozens that is not significant. I've seen what looked great systems and methodologies in backtests have more weakness when going forward because of the inherent flaws in backtesting and insufficient data points. Thank you for your work.

A: The number of data points typically averages between 60 to 90 round trip trades per year going back to January 2009, more than 8 years of data. I have also spot checked going back to 1999. 

Backtests cover a series of both bull and bear market cycles, choosing "cream of the crop" results has never been the case, profits are balanced (not one-time), and any adjustments have an underlying logic which is proven out by many years of data.

The backtests eliminate discretion and judgement except during rare times of higher or spiking volatility when the strategy has a certain level of profit on a buy signal, typically in the double digit percentages. I then gauge price/volume action to see where I would most likely take profits. I generally am looking to exit at a minimum profit of at least 10-20%, though there are rare exceptions where I believe much larger profits await, thus may decide to hold onto the position longer than usual, with the downside of reversing the gains and ending at a loss.

I assume 0.2% slippage with each trade even though in practice, slippage has been less.

When building a strategy, one must guard against taking a set of parameters that worked well in the past as they are likely to fail going forward if the parameters were overfit to past data. They would thus have no predictive ability going forward.

You can test this theory very quickly in your own strategies by assigning parameters that have been optimized over a period of years, say from 2000-2009 (in-sample data), then seeing if they work in 2010 (out-of-sample data). My guess is they would fail if the parameters were over-optimized to fit past data.

When building a strategy, each rule must have inherent logic backing it. And should markets change in a material way, any changes introduced into the model must continue to contain inherent logic, otherwise one may be overfitting using that particular parameter.

Another factor to guard against is big profits on just a few signals since outliers should be rejected. A Q-test insures against this, though as one becomes comfortable with data analysis, it becomes very simple to spot.

The adjustment I made starting with the 4-3-17 sell signal had been tested with both in-sample and out-of-sample data (real-time). The near record down day in the VIX that gapped lower overnight does not invalidate the adjustment. I built the strategy to be beyond 6-sigma resilient since black swans do happen more often than we realize.

That said, I am closely monitoring the strategy's risk levels before and after the adjustment.

The motivation behind the adjustment is a substantial boost in reward with fewer whipsaws. The downside is fail-safes that allow for somewhat greater loss, though over time, the reward should well outperform risk.

Published: Apr 25 2017, Modified: Apr 30 2017