EDUCATION: Bitcoin's Recent Surge and the Problem w/ Williams %R
It's now more than two weeks since my successful call of Bitcoin testing the bottom, with the BTC price increasing more than $2K since my initial idea was published. The price action hasn't been too volatile since a 1K spike 10 days ago, something which caught my attention as well.
Despite this successful run from BTC, the continuous indication of "overbuying" per Williams %R has been problematic for my analysis. The goal of my published ideas is to indicate outliers that predict upcoming regression, which in turn presents ideal BUY and SELL points for Bitcoin and other cryptocurrencies. That goal has been recently compromised by %R indicating an overbuy for 11 of the past 12 days. While this indicator puts us on alert to sell, there isn't any actual market substance behind the call to action. And thus, the problem with Williams %R is illustrated.
Allow me to explain, but first, let's look at the equation for %R:
%R = (Highest High - Close)/(Highest High - Lowest Low) * -100
Any security with a %R over negative-20 is considered to be overbought, while any under negative-80 is oversold. In other words, applying current context, if the distance from the current BTC price to the 14-day high is less than 1/5 of the 14-day high-low range, Bitcoin is overbought. So in theory, the 14-day high (and anything 5+ times smaller than the 14-day range) will ALWAYS be overbought. This is far too simplistic for the stat to be constantly relied upon for day trading. The results are NOT distributed normally enough for the indicator to consistently find the proper outliers.
Just look at BTC. Bitcoin dropped from 19K to 6K mostly due to regression from the massive buy outlier that got the coin to its all-time high. BTC then bottomed out around 6K because it was pushing the boundaries of normality, per the regression trendline established in my first published idea. The current push back near 9K is natural as well. So why should any 14-day peak in this context be looked at as a legitimate overbuy, especially when the BTC price action is arguably the most controlled its ever been?
Also, look at %R from a structural standpoint. It is far too simplistic, compared to the likes of RSI. The averaging required for the RSI helps to normalize the result distribution. This is how RSI is calculated:
RSI = 100 - (100 /(1 + R.S))... R.S = Average Gain / Average Loss
Note that an RSI above 70 is considered overbought, while an RSI under 30 is considered oversold. In other words, the average gain would need to be more than 7/3 times greater than the average loss over a 14-day span for Bitcoin to be overbought. Here, we see averages upon averages.
The gains and losses for RSI are calculated from close to close in my ideas, meaning that absolute highs and lows over the 14-day period are not considered. Here's the important thought that comes to me: statistical indicators should be designed to locate outliers, and as such, outliers should be eliminated with the equation itself. I'll illustrate my thought more simply in a sports context. Would you rate players' performance based on the success of Michael Jordan or LeBron James, as opposed to average-level players? Of course not, because most players look inferior compared to GOAT-tier players. For this reason, the Williams %R is prone to providing misleading information in certain contexts.
As mentioned in one of the comment updates of my last idea, I will drop %R from my analysis. While sometimes it is a great indicator, it is too flimsy in application for me to constantly trust it. I'd much rather use RSI, which distributes its results more normally than %R.
I hope this education has helped in some way. My best to you, as we trade!
Despite this successful run from BTC, the continuous indication of "overbuying" per Williams %R has been problematic for my analysis. The goal of my published ideas is to indicate outliers that predict upcoming regression, which in turn presents ideal BUY and SELL points for Bitcoin and other cryptocurrencies. That goal has been recently compromised by %R indicating an overbuy for 11 of the past 12 days. While this indicator puts us on alert to sell, there isn't any actual market substance behind the call to action. And thus, the problem with Williams %R is illustrated.
Allow me to explain, but first, let's look at the equation for %R:
%R = (Highest High - Close)/(Highest High - Lowest Low) * -100
Any security with a %R over negative-20 is considered to be overbought, while any under negative-80 is oversold. In other words, applying current context, if the distance from the current BTC price to the 14-day high is less than 1/5 of the 14-day high-low range, Bitcoin is overbought. So in theory, the 14-day high (and anything 5+ times smaller than the 14-day range) will ALWAYS be overbought. This is far too simplistic for the stat to be constantly relied upon for day trading. The results are NOT distributed normally enough for the indicator to consistently find the proper outliers.
Just look at BTC. Bitcoin dropped from 19K to 6K mostly due to regression from the massive buy outlier that got the coin to its all-time high. BTC then bottomed out around 6K because it was pushing the boundaries of normality, per the regression trendline established in my first published idea. The current push back near 9K is natural as well. So why should any 14-day peak in this context be looked at as a legitimate overbuy, especially when the BTC price action is arguably the most controlled its ever been?
Also, look at %R from a structural standpoint. It is far too simplistic, compared to the likes of RSI. The averaging required for the RSI helps to normalize the result distribution. This is how RSI is calculated:
RSI = 100 - (100 /(1 + R.S))... R.S = Average Gain / Average Loss
Note that an RSI above 70 is considered overbought, while an RSI under 30 is considered oversold. In other words, the average gain would need to be more than 7/3 times greater than the average loss over a 14-day span for Bitcoin to be overbought. Here, we see averages upon averages.
The gains and losses for RSI are calculated from close to close in my ideas, meaning that absolute highs and lows over the 14-day period are not considered. Here's the important thought that comes to me: statistical indicators should be designed to locate outliers, and as such, outliers should be eliminated with the equation itself. I'll illustrate my thought more simply in a sports context. Would you rate players' performance based on the success of Michael Jordan or LeBron James, as opposed to average-level players? Of course not, because most players look inferior compared to GOAT-tier players. For this reason, the Williams %R is prone to providing misleading information in certain contexts.
As mentioned in one of the comment updates of my last idea, I will drop %R from my analysis. While sometimes it is a great indicator, it is too flimsy in application for me to constantly trust it. I'd much rather use RSI, which distributes its results more normally than %R.
I hope this education has helped in some way. My best to you, as we trade!
RSI is calculated with the help of smoothing, in that the average gain and average loss used for the equation accounts for previous average gains and average losses in order to get its results. This smoothing technique helps to improve accuracy as the overall collection of data increases. So basically, you need the daily data for Coinbase's Bitcoin trading since Day 1 to have the most accurate 14-day RSI.
For more info, check out this explanation of RSI: stockcharts.com...school/doku.php?id=chart_s...
This explanation can provide more clear answers than I can type here in limited space. So it's best to read all of it to have a better understanding on how RSI very effectively notes outliers that are overbought or oversold.