Analyzing Trading Indicators for Cryptocurrencies

Modified from Photo by Kevin Ku from Pexels https://www.pexels.com/photo/coding-computer-data-depth-of-field-577585/

TL;DR: Utilizing hourly data, instead of daily data, while analyzing technical indicator signals for cryptocurrency trading appears to have some value for a select few coins, but not all.

We concluded in our last article that technical indicators, when applied in isolation, tend to be ineffective at predicting the types of future returns that analysts may be looking for. In this article, we apply our approach to more granular data and see if this result still holds.

A more detailed overview of our process exists in the article “Do Technical Indicators Really Work on Cryptocurrencies?”. But for those with not enough time to read through it, here is a quick overview of the findings followed by the detail.

Summary of Findings

Lower bollinger band crossovers look promising as an upside predictor in general.
VeChain still looks like an outlier, ignore it for technical analysis.
When it works, predictive power seems short-lived, probably not useful for more than 12 hours.
DASH seems to react well to RSI and MFI downside indicators consistently over many time periods.
Upper bollinger crossovers seem to work well for a few coins (EOS, LSK, VEN, and ZEC) but not consistently across all coins.

Analysis

If you’ve made it this far, you must be interested in how this analysis works. Keep reading to learn more.

Method

We select the top 20 cryptocurrencies based on market cap as listed on CoinMarketCap.com.
We select an analysis time period and disqualify any coins that do not have enough data to produce meaningful results within that time period. In this case, we chose the analysis period of September 1st, 2017 through February 8th (the date of this first draft). This time frame was chosen due to its recency as well as both the positive and negative returns observed for most cryptocurrencies during this time period.
We define a set of commonly used technical indicators with typical cross over event periods 20/80 for MFI, 30/70 for RSI, etc.
We analyze the number of occurrences of of these indicator events to see if they are worth analyzing. If something only happens a handful of times, there is probably not enough statistical power there.
We define success criteria for both positive and negative indicators in order to measure their effectiveness. For indicators that supposedly predict price increases, we measure how much higher than average returns were achieved. For indicators that supposedly predict price decreases, we measure how far below zero the future prices were.
The analysis in step 5 is repeated for multiple time periods after the event occurs: one, six, twelve, and twenty-four hours are used in this analysis.

The analysis was performed using a combination of the Gatsiva API and Python / MatPlotLib for visualization.

Coins Analyzed

The following coins met the criteria for both market cap and enough data during our test period.

Table 1. Coins Analyzed

Occurrence Analysis

Because this is hourly data, the occurrence rates for most of these rules was significantly higher than our daily data analysis. Below are heatmaps of occurrence rates for technical indicator events by currencies.

Downside / Upside Event Occurrence Comparisons

From these charts we can make few observations:

We can see that ema/sma crossovers on 50/200 are relatively rare events for most currencies.
Additionally, we notice mfi/rsi crossovers occur much less freqently than macd histogram, stochastic, and bollinger band ranges.
Finally, we see relatively the same pattern across all analyzed currencies.

Now let’s move on and take a look at the actual results. Remember when looking at the following charts, we have defined the following approach for scoring these technical indicator events.

Table 2. Scoring Approach

Analyzing Upside Results

The following charts show the score of each technical indicator event for each currency as a heatmap. Green scores indicate good performance. In this case, good performance is an average return that exceeds the regular average return otherwise observed if we did not pay attention to the signal.

Upside performance results for 1, 6, 12, and 24 hours after event

From these charts, we can make the following observations:

“bollinger range crossing below 0” looks promising, but again, we need to remember we have been mostly in an upside only return environment.
VeChain still looks like an outlier when it comes to technical analysis. Though it looks like it does well in the upside analysis, you will see in the downside analysis that it does quite poorly.
A few coins appear to react well to sma(50) / sma(200) cross up including BTC, DASH, EOS, ETC, and XLM. But we need to temper that with a recognition that these events happen much less frequently.
There are many more positive results after 1 hour than after 6 hours, suggesting that there may be some degree of predictive power, but it could be short-lived.

Analyzing Downside Results

Ok, now let’s take a look at the downside technical indicators. Remember in this case, positive return performance is marked as red while negative return performance is marked in green. This is because downside indicators are supposed to predict downward price movements.

Downside performance results for 1, 6, 12, and 24 hours after event

From these charts we can make the following observations:

DASH seems to react well to RSI and MFI events consistently over many time periods.
Bollinger Range tops seem to work well for a few coins (EOS, LSK, VEN, and ZEC) but not consistently across all coins.
Most effects seem short-lived as the returns turn positive (bad when predicting downside results) within 24 hours.
VEN is a hot mess and really doesn’t respond the same as other cryptocurrencies.

Future Analysis

We always like to follow-up our articles with what is next, or what we have missed. Some observations we have made internally include:

Using ranges instead of crossovers: We are using simple crossovers, ranges of events might be more appropriate in this market. Fortunately, the Gatsiva API now supports within x% of type indicators to allow for this to happen.
Tuning indicator time periods: As mentioned in our prior article, we know from experience that standard time periods for technical indicators used in equities require tweaking when used for cryptocurrencies. Our genetic algorithm research has proven this. Exploring which parameters work best for these indicators would be a worthwhile exercise.
Looking at how these returns change over time: This is what Gatsiva does already with our confidence band shift analysis. By looking at the trends of indicators over time, we have the ability to see if technical indicator is trending as a better or worse indicator.
Applying a different definition of “predictability”: There is more than one way to look at predictability. Options could include using median instead of average, looking for a minimum or maximum threshold of returns, etc.

About Gatsiva

At Gatsiva, we provide APIs and tools empowering analysts to do similar research to what’s been presented above. We also provide research and education articles that help traders and analysts determine the viability of technical signals.

In addition, Gatsiva uses machine learning and genetic algorithms to find the technical indicator events that actually work, and track their performance over time. We believe this is a strong differentiator when compared to using common indicators such as those provided in this analysis.

To learn more, you can visit us at https://gatsiva.com or follow our analysis at https://twitter.com/gatsiva.

Finally, if you like what you just read: Feel free to bookmark this article, clap, or leave a comment! It helps us help more people!