Keltner Channel Study: Mean Reversion

I. Introduction & Overview

II. SPY Results

III. VIX Results

IV. Conclusion

I. Introduction & Overview

The purpose of this test is to evaluate the usage of volatility bands in measuring short-term overbought and oversold conditions, as well as find their most optimal settings. Keltner Channels under certain settings have been shown to be effective in a statistical analysis conducted by Adam H. Grimes (https://adamhgrimes.com/library/indicators/keltner-statistics/). This study will expand on Grimes' findings, particularly his "Keltner Fade" strategy, which will be explained later on.

To introduce Keltner Channels, they are a similar indicator to the popular Bollinger Bands in function, but differ in calculation. While Bollinger Bands rely on standard deviation of price, Keltner Channels are calculated using Average True Range (ATR), which is a measurement of the average distance a security has moved over a certain period of time. Like most indicators, Keltner Channels can take a variety of different settings, including Moving Average type and periods, and ATR type and multiplier. In his statistical study, Adam H. Grimes suggests general usage of Keltner Channels with the following settings:

- 20 Exponential Moving Average (EMA)

- Simple ATR

- 2.25 ATR multiplier

The 20 EMA is plotted, where the upper and lower bands are then calculated by adding/subtracting 2.25*ATR. A visualization of Keltner Channels under these settings are shown below:

In theory and by observation, rapid price movements above the upper band and below the lower band can help signal potential mean reversion trade opportunities. In this backtest, any closing below the lower Keltner band will be bought (referred to as fading long) and any closing above the upper Keltner band will be shorted (referred to as fading short). A variety of ATR multiplier settings will be tested, using Grimes' 2.25 ATR multiplier as the baseline. Results for holding the trade for 1, 2, 3, 4, 5, 10, 15, and 20 days will then be reported. A visual example of the test conditions is shown below:

In this study, the SPDR S&P 500 ETF Trust (SPY) and the CBOE Volatility Index (VIX) will be tested on both daily and weekly timeframes and report the most optimal ATR multiplier for each fade strategy. This backtest will begin on 01/01/2003.

II. SPY Results

When buying any close below the lower Keltner band (fading long) on a daily timeframe on the SPDR S&P 500 ETF Trust (SPY), a 2.00 ATR multiplier yielded the best results and following summary statistics:

Keltner Fade Long - Daily (ATR Multiplier: 2.00)

When shorting any close above the upper Keltner band (fading short) on a daily timeframe, Grimes' 2.25 ATR multiplier yielded the best results and following summary statistics:

SPY Keltner Fade Short - Daily (ATR Multiplier: 2.25)

It is significant that all mean returns were positive and the lowest win frequency reported was 62.6%. Although the results when fading short were not as impressive, there is still some significance in positive mean returns in the first 5 days of holding. With positive mean returns but higher lose frequencies, it is inferred that successful trades were larger than unsuccessful trades, offsetting the higher lose frequency. However, less significant results suggests that the decision to fade short should be supported by other analyses. When fading long, mean returns and win frequencies generally tended to increase as time progressed and the position was held. The opposite is true up to a certain point when fading short. This is likely due to the S&P 500's strong bullish tendency.

Theoretically, the effect of randomness should be lessened as timeframe increases. Therefore, as expected, the same test on weekly timeframes yielded better, steadier results. Unfortunately, price closing below the lower Keltner band did not occur enough to garner a sufficient sample size to evaluate the fade-long strategy. In an attempt to compensate, the ATR multiplier was decreased to 1.50 to bring the sample size up to 20. When fading long on a weekly timeframe, a 1.50 ATR multiplier yielded the following summary statistics:

SPY Keltner Fade Long - Weekly (ATR Multiplier: 1.50)

When fading short on a weekly timeframe, Grimes' 2.25 ATR multiplier once again yielded the best results and following summary statistics:

SPY Keltner Fade Short - Weekly (ATR Multiplier: 2.25)

With these fantastic results, it is shown that there is a greater edge in "fading" by performing mean reversion trades on a weekly timeframe than on a daily timeframe. This suggests that price closing above or below the Keltner bands on a weekly timeframe, rather than a daily timeframe, could have more merit in measuring overextension and predicting future mean reversion. Additionally, better results and higher probabilities may be presented when overextension is supported by multiple timeframes at once.

Due to the sample size issue when fading long, determining the most optimal ATR multiplier setting for fading long on a weekly timeframe cannot be concluded. Therefore, it is recommended that traders choose a setting in anywhere between 1.50 and 2.25. The setting of choice may not have a significant impact because in the rare occurrence that price closes below the lower Keltner band on a weekly timeframe, traders will likely fix their attention towards the underlying fundamental, macroeconomic reasons behind such a drop rather than consult a technical indicator. Therefore, to present the most optimal ATR multipliers for each strategy and timeframe clearly:

These settings were also tested on a variety of other symbols, including those who do not trend as strongly as the SPY, to ensure consistency; the conclusion stands.

If a trader is able to use different ATR multipliers for the Keltner Channels' lower and upper bands on their price charts, the statistics reveal they should consider doing so. For those who use TD Ameritrade's ThinkOrSwim trading platform, it is possible to adjust the ThinkScript code to allow different multipliers for the upper and lower Keltner band. You can also use import my code adjustments for ThinkOrSwim's Keltner Channels here:

http://tos.mx/bjT07VN

III. VIX Results

When fading long on a daily timeframe on the CBOE Volatility Index (VIX), a 2.00 ATR multiplier yielded the best results and following summary statistics:

VIX Keltner Fade Long - Daily (ATR Multiplier: 2.00)

When fading short on a daily timeframe, a 2.00 ATR multiplier also yielded the best results and following summary statistics:

VIX Keltner Fade Short - Daily (ATR Multiplier: 2.00)

It is clear that fading with Keltner Channels using a 2.00 ATR multiplier yields incredible results on the VIX. Traders can clearly profit from quick or "overshot" changes in volatility, especially when volatility spikes and is shorted. However, since the VIX itself cannot be bought, traders will need to trade either a VIX-priced instrument such as the UVXY or VXX, or VIX derivatives, which are priced off of VIX Futures (/VX). In these scenarios, traders should be aware of potential lag or deviation from the actual VIX symbol.

IV. Conclusion

Volatility bands can be powerful tools in gauging short-term overbought and oversold conditions in equities. This is revealed by the consistent positive mean returns reported when using the most optimal Keltner Channel settings. Proper use of volatility bands can help signal potential mean reversion trading opportunities. Similar statistical work may also be conducted on Bollinger Bands.