Cryptocurrency Technical Analysis: Is it Effective?

The advent of cryptocurrencies has given rise to a new wave of aspiring traders that use technical analysis, mostly in its classical form, in an effort to profit from those markets. But is technical analysis effective in high volatility markets? In this article we look at some data and try to provide few answers.

What is technical analysis?

I have written in the past that traditional technical analysis is a form of lossy compression that maps prices to visual representations, for example chart patterns, trendlines, indicators, etc. With lossy compression, as the name implies, information about price action is lost for the benefit of obtaining an “alphabet”, or representation, of the data that can be used in forecasting.

One of the main problems of technical analysis is that the effectiveness of the method was not quantified when it was first published in books, mainly in mid 20th century, but was presented to traders in the form of indisputable truths some masters of the art decided to reveal. There were no backtests with success rate, expectation, payoff and Sharpe ratio of those classical methods of technical analysis because one of the problems was that computing was still in its early phases. The published books, some of which are even used nowadays, included mostly examples of pattern formations that worked well in the past without any reference to their statistical properties. But careful analysis would reveal that although some of these patterns, such as the flagship head and shoulders, may have worked in the past, their mathematical expectation (amount won or lost on the average in the longer-term as samples become sufficient) has fallen to zero or has even turned negative in recent years.

A head and shoulders is an easy pattern to spot while the advent of the Internet has facilitated the dissemination of information related to these chart patterns. It is impossible for a large number of traders that use these patterns to profit and as a result profitability has reverted towards zero in the information age. When a potential edge becomes widely known, it naturally ceases to be an edge. But was there an edge in the first place?

This is an interesting question in my opinion and I have tried to answer this in my articles and in my book, Fooled By Technical Analysis, in the following context:

Technical analysis was developed mainly for the equity markets (futures and currency markets did not even exist in the early 20th century) during a period of high serial correlation of daily returns, as shown in the chart of Dow Jones Industrial Average below since 1940. (Click on image to enlarge.)

Figure 1. DJIA daily chart with 1-lag, 252-day autocorrelation of daily returns. Chart created with Amibroker,

A crucial regime shift

Other than for a brief period in the early 1960s, the 1-lag, 252-day autocorrelation of daily Dow Jones Industrial Average returns was high and statistically significant until the mid 1990s (it rose above the upper significance band at 0.123, as shown in Figure 1.) In the early 1970s, the autocorrelation peaked and then started decaying. This was also the case will other equity indexes, such as the S&P 500. During that time early quant traders emerged, i.e., traders that started to use computers for forecasting market prices. By mid 1990s, the high autocorrelation was arbitraged out of the market and prices became mean-reverting with autocorrelation turning significantly negative. That was a major market regime shift that neutralized classical technical analysis. Here is why:

Most classical analysis patterns depend on the notion of confirmation: prices must rise or drop below a certain level before a pattern is confirmed. With high serial correlation, confirmation was easier because, by definition, there was high probability of up days followed by up days and down days followed by down days. In essence, it is highly probable that the pioneers of technical analysis were fooled by the high autocorrelation and attributed the momentum in price action to a predictive power of some chart formations. But when the regime shift occurred in late 1990s, this assumed predictive power vanished. In the case of a head and shoulders for example, prices after dropping below the neckline could reverse and invalidate the pattern due to mean-reversion, before hitting the expected profit target.

Below is an example from Russell 2000 index, included in the book.

Figure 2. Head and shoulders pattern failure in Russell 2000. Source: Fooled By Technical Analysis Chart created with Amibroker,

The head and shoulders pattern in Figure 2 took about three months to form but after confirmation, prices reversed causing invalidation. This type of failure is typical with most classical technical analysis formations that are widely advertised in blogs and social media and it is also natural because the market cannot serve the appetite of profit of all aspiring traders that use them and at the end of the day some, if not most of them, must lose. These patterns are obvious to anyone with access to charts and for this reason they provide no edge.

In fact, one could find head and shoulders formations in other markets that have generated profits but in the longer-term, the expectation from trading these patterns in mean-reverting markets is zero before commission cost is included. While some old-timers insist they can profit from these patterns if they make a lot more then they lose on the average even if their win rate is low, such claims are naive and fail simple mathematical verification. Since these patterns take several months to form on the average, trading a sufficient sample of them to allow determining their expectation with high statistical significance would require a few years and during that period risk of uncle point is high. More details along with the math can be found in the article The Patient Chart Pattern Trader.

In a nutshell, there are two factors that limit the profitability of technical analysis: (1) mean-reversion and (2) high volatility. Below we explore these two factors in the case of bitcoin and ethereum.


Below is a bitcoin daily chart (semi-log scale) with the 1-lag, 365-day autocorrelation of daily returns.

Figure 3. Bitcoin daily chart with 1-lag, 365-day autocorrelation of daily returns. Chart created with Amibroker,

The autocorrelation pattern is clear: after the 2014 rise and subsequent collapse, a regime shift from momentum to mean-reversion occurred and it was quite significant. After that shift, this market exhibits mean-reversion. Note that uptrends and downtrends can form in mean-reversion regimes but they are more difficult and more risky to trade.

In ethereum the autocorrelation pattern is also interesting, as shown in Figure 3.1 below.

Figure 3.1 Ethereum daily chart with 1-lag, 365-day autocorrelation of daily returns. Chart created with Amibroker,

This cryptocurrency started in mean-reverting mode and then there was an increase in autocorrelation that ended last June. This market has also entered a mean-reverting regime.

One of the main mechanisms that enforces a regime swift from momentum to mean-reversion is algo trading. Algos arbitrage autocorrelation and make markets more efficient. In effect, algos limit the profit potential of speculators that attempt to use naive compression schemes, such as technical analysis. This is good for price discovery but it is bad for speculators.

Our Momersion indicator that measures the relative magnitude of momentum versus mean-reversion provides additional information. Figure 4 shows the percentage of up days in a 365-day rolling window that are followed by up days and the percentage of down days in the same rolling window that are followed by down days, for bitcoin.

Figure 4. Bitcoin daily chart with UP-UP and DOWN-DOWN Momersion indicator. Chart created with Amibroker,

It may be seen from Figure 4 that in the beginning of this year approximately 60% of up days in a 365-day period were followed by up days while about 40% of down days were followed by down days. This dynamic supported the explosive uptrend that suddenly ended with a crash. As of August 18, only 51% of up days in a 365-period are followed by up days while 46.86% of down days are followed by down days. Essentially, the bitcoin market is close to random at this point from this perspective.


The volatility of cryptocurrency markets is very high due to their speculative nature. Below is a chart of bitcoin with the 30-day standard deviation (also annualized) and the 14-day Average True Range (ATR) as a percentage of closing price.

Figure 5. Bitcoin daily chart with volatility indicators. Chart created with Amibroker,

The 30-day standard deviation was 3.28% as of August 18, 2018. Annualized for 365 days, the same standard deviation was at 62.76%. The 14-day ATR was at 5.1% of the closing price. Notice how the 30-day standard deviation reached as high of 8% in the beginning of this year and annualized at a high of more than 160%, while the 14-day ATR climbed to a high of 20%.

The above numbers are so high that rule out profitability of conventional trading methods, such as technical analysis. How high are they? Let us compare them to the same volatility measures in of SPY, the S&P 500 ETF, shown in Figure 6.

Figure 6. SPY ETF daily chart with volatility indicators. Chart created with Amibroker,

The 30-day standard deviation for SPY is 0.49%, annualized for 365 days it is at 9.5% and the 14-day ATR is at 0.78%. These risk metrics in the case of SPY are one order of magnitude (a factor of 10) lower than those of bitcoin and this is a huge difference. If trading SPY is difficult to start with, the high volatility makes cryptocurrency trading extremely difficult and highly unprofitable for most participants.

We have determined that most technical analysis indicators are not profitable in cryptocurrencies. We use a proprietary indicator in our premium reports that attempts to time overbought/oversold conditions, called PAL OB/OS, in the weekly timeframe because in the daily timeframe the signal to noise ratio is very low. Below is a chart that shows the recent performance of this indicator.

Figure 7. Bitcoin daily chart with PAL OB/OS indicator. Chart created with Amibroker,

This indicator worked well in identifying the major top last year and also the recent correction. It has also worked well in timing bottoms. However, we use this indicator in conjunction with other analysis. Overbought/oversold indicators usually measure uptrend/downtrend momentum and their use in a contrarian sense carries high risk because these conditions can persist for extended periods of time. It takes significant experience with price action and skill to use indicators of this kind profitably. They are not recommended for use by aspiring cryptocurrency traders.

What methods offer potential to aspiring cryptocurrency traders?

In our opinion, cryptocurrency traders should avoid gurus, educators and old timers that recommend classical technical analysis and concentrate on sharpening their quant skills. Starting with a small capital and exercising prudent risk and money management are good ways of learning the market and increase chances of profitability. There are also some good platforms that allow paper trading for getting familiar with execution mechanics. The road to profitability, especially in high volatility markets, is filled with traps. There is no method to guarantee windfall profits and any such claims are either due to luck or unusual skill and cannot be replicated. Since this is a new market, not too many have experience with it and any claims of expertise should be carefully scrutinized. Finally, for most people that believe in the potential of cryptocurrencies as a solution to fiat money problems, the best solution is to avoid trading and just hold. If you are not sure of what to do, you should contact a competent financial adviser. Trading cryptocurrencies involves substantial risks and there is high probability of total loss.

About the author

Michael Harris is a trader and best selling author. He is also the developer of the first commercial software for identifying parameter-less patterns and related anomalies in price action 17 years ago. In the last seven years he has worked on the development of DLPAL, a software program that can be used to identify short-term anomalies in market data for use with fixed and machine learning models. Click here for more.


No part of this article constitutes an offer to buy or sell any sort of security or financial product. Furthermore, no part of this article should be construed as advice or a recommendation for any type of financial transaction or investment decision. For our full disclaimer click here.

© 2018 Michael Harris. All Rights Reserved

This article originally appeared in Price Action lab blog.