Photo by energepic.com from Pexels TL;DR: Generally used technical indicator events do NOT appear to be good predictors of above average, or negative, returns for cryptocurrencies when utilizing daily prices and when used alone. In addition, in markets that have moved primarily upward, it is probably unwise to put too much faith in indicators that predict upside performance. Our Hypothesis The hypothesis is that technical indicators are untested in this market, and as a result, analysts and traders may or may not be applying them properly. We set out to find out whether or not certain technical indicators have any value in predicting any upside or downside performance. Our Process We define a set of cryptocurrencies to analyze, using market cap and amount of available data as our filtering criteria. We define a set of technical indicator events to test. Note that these indicators are not meant to be representative of all technical indicators, but rather a worthy, and frequently used few with which to illustrate our analysis. (Note: If you have additional indicators you’d like to see in this analysis, please let us know in the comments below). We define what “predictive value” means for this exercise and analyze the returns after various events utilizing this criteria. 1. Defining the Cryptocurrencies To Analyze First we define which currencies we want to look at. We can start by looking at the market cap of each coin on . For this exercise, we haven chosen the top 20 coins by market cap and removed any coin that doesn’t have more than 2 years of daily data. This leaves us with the following coins to analyze. Coin Market Cap Bitcoin (BTC) (ETH) Ethereum Ripple (XRP) Litecoin (LTC) Dash (DASH) VeChain (VEN) Monero (XMR) It is worth noting that many of the more popular coins such as Cash (BCH), Cardano (ADA), Neo (NEO), etc. have less than 2 years of data. As a result they are currently removed from this analysis. Bitcoin Two years of data represents approximately 730 data points, since crypto trades every date at all times. Some of our indicators, like simple moving average over 200 periods, require at least that many periods to be relevant. This gets even more complicated when you begin bootstrapping exponential averages for other indicators like RSI and Stochastics. Therefore, having a good set of data to begin with that will cover the priming period of many indicators is essential. Why do you need two years of data? 2. Defining the Technical Indicator Events Probably the most important aspect of this research is choosing which technical indicator events to analyze. To start this research, we chose a handful of commonly used technical indicator events and evaluated both the cross above and cross below events to try to capture directionality. For example, “RSI below 30” is often cited as a “oversold” indicator, indicating that the price should rise from that point forward. To capture this event, we use both entering that condition and exiting that condition . rsi(14) crosses below 30 rsi(14) crosses above 30 Many of the technical indicators are traditionally utilized with “default” values. We have noticed that traders are tending to use these default values in analyzing cryptocurrencies as well. Though Gatsiva does this analysis internally via our genetic algorithms, “tuning” these parameters for a particular currency is beyond the scope of this article. Why choose the value inputs of **14** for RSI, or **12,26,9** for MACD? In addition, we will also divide our events into two categories for analysis purposes: and events. Downside events will be those that typically predict an overbought condition and are supposed to predict future lower prices. Upside events will be those that typically predict oversold conditions and should be indicators of future higher prices. downside upside 3. Defining “Predictive Value” For our analysis, we will define the “score” of a technical indicator event by how well it performs on average after an event occurs. Events that are supposed to predict upside performance should produce returns above the average return. Events that are supposed to predict downside results should see a negative, or at worst, a neutral return. Therefore, we will use the following approach to “score” our events: Table 1. Event Type and Score Definitions : We recognize analysis could certainly be repeated using various criteria for “predictive value”. If you have ideas, please leave them in the comments below. Note Finally, since we don’t really know what periods are worth looking at, we’ll compare returns at the 1, 7, 14, and 28 day intervals after an event occurs. The Analysis We’ve written this analysis in Python script, but it could be done in any language by utilizing the Gatsiva API to retrieve market data and retrieve the list of dates for which a particular event is true. For more information please see the and transactions of our API documentation. Condition History Symbol Price Data Looking at Occurrences The first thing we want to do is check out how often these events actually occur during our test period. Since we selected our coins by the criteria of who had at least 2 years of data, we selected 01–01–2016 through 02–11–2018 as our period of analysis. We can then easily check how often these events occurred for each coin. The following charts below shows a heat map of occurrences by coin and by event, separated by downside and upside events. Figure 1. Upside Event Occurrences Figure 2. Downside Event Occurrences Observations: The SMA and EMA 50/200 crossover events happened very rarely for all coins. These events are probably not useful to predict performance since they happen infrequently. VeChain (VEN) had very few technical indicator events occur in general. Therefore, we will remove it from our analysis moving forward. In general, events that occur when prices are high tend to occur more often (e.g. bollinger range breaching 1, full stochastics crossing above or below 80, rsi crossing above and below 70, and mfi crossing above and below 80). This is not surprising since cryptocurrencies in general have been increasing rapidly over the past 2 years. Looking at Baseline Returns Next we calculate the average return after 1, 7, 14, and 28 days. For this exercise, we are simply averaging the returns observed after every period. Here we can see that every coin within our scope of analysis has gone up. Table 2. Average Returns For 1, 7, 14, and 28-day Periods This table is an important set of facts to keep in mind as we analyze post-event returns. Many upside technical indicators will appear to work simply because the entire trend of prices is upward. However, we can adjust for this as mentioned above by defining “predictability” for upside events as the extent to which they that is otherwise observed. exceed the average return Calculating and Visualizing “Predictive Value” Now comes the main core of our analysis. To start we will look at the return the period immediately after the event occurs, in this case, 1 period afterwards. In the case of upside events, green cells indicate areas of stronger than average performance, red cells indicate areas of weaker than average performance. In the case of downside events, as indicated before, green cells indicate areas of negative performance (what we would want to see from a downside event), and red indicates areas of positive performance (what we would want to avoid with a good downside event). So remember, in both cases here, green is good! Areas in grey represent areas with no results because there were no events observed and thus, no score available. Upside Event Observations: Figure 3. Upside Event Returns - 1 Day After Event For upside events we see very inconsistent performance. With the exception of potentially , which only showed negative performance for DASH. There are no events which consistently show better than average positive performance for multiple coins. rsi(14) crosses below 30 and show interesting potential performance for BTC on a daily basis. However, both of these events occurred quite infrequently (12 times and 4 times respectively) over the last 2 years. full stochastics d crosses below 20 rsi(14) crosses below 30 show interesting potential performance for DASH, XMR, and XRP on a daily basis. However, this events occurred less than 5 times over the last 2 years. For LTC, this event occurred 7 times but the performance is not above average. sma(50) crosses above sma(200) In addition, we see quite a bit of negative performance as well spread throughout other currencies, indicating that these events don’t seem to predict any general results. Downside Event Observations: Figure 4. Downside Event Returns — 1 Day After Event Overall, we see very little good performance (Remember: in this case, good performance would be a negative return). This is likely because of the huge run up in value during the 2017–2018 time period. ema crossovers appear to have some value for ETH. However with only 7 observations of the ema 10/20 crossover and only 1 observation of the ema 50/200 crossover, we wouldn’t put much weight behind it at the moment. There are some other very small pockets of good performance, but overall we see inconsistency between coins like we observed in the upside event results. Now what we can do is look at other periods. The following images show the upside and downside event performance for 7, 14, and 28 days after the events. 7 Days After Event Figure 5. Upside Event Returns — 7 Days After Event Figure 6. Downside Event Returns — 7 Days After Event 14 Days After Event Figure 7. Upside Event Returns — 14 Days After Event Figure 8. Downside Event Returns — 14 Days After Event 28 Days After Event Figure 9. Upside Event Returns — 28 Days After Event Figure 10. Downside Event Returns — 28 Days After Event In general we see the exact same pattern with one exception that could be explored further. BTC seems to show relatively good performance with upside events, but only after 28 days. We suspect that this is because of the strong upward momentum of BTC prices over the history of the analysis period and has little to do with the specific indicator event in question. Conclusions Based on daily data over the last two years, it we have chosen for this analysis when utilized on daily data in a standalone fashion. does not appear that the technical indicator events have strong predictive value for finding above average or negative returns Future Analysis Though these indicators may not appear to work, we don’t think this is necessarily the end of the road. Some potential paths for follow-up research include: We know we only chose a small subset of indicators. There may vbe others out there that work much better for cryptocurrencies. Expanding the set of indicators: Indicator events are often used to support a thesis, not used alone. Looking at combinations of signals may be fruitful. Looking at combinatorial aspects of indicators: We know from experience that standard time periods for technical indicators used in equities require tweaking when used for cryptocurrencies. Our has proven this. Exploring which parameters work best would be a worth exercise. Tuning indicator time periods: genetic algorithm research This is what Gatsiva does already with our . 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. Looking at how these returns change over time: confidence band shift analysis There is more than one way to look at predictability. Options could include using median instead of average, looking for a , etc. Applying a different definition of “predictability”: minimum or maximum threshold of returns It is possible that trends for daily data may be too volatile, but trends for hourly might yield results. We are working on this analysis now and will publish Part 2 shortly. Changing the timescale of analysis: 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 or at . https://gatsiva.com 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!
