Weekly Cyclicality in Cryptocurrency Prices

A look at price patterns per day-of-the-week in Bitcoin, Ethereum, and Litecoin. On deck: the performance of trading strategies based on the weekly cyclicality analyzed here.

When my first intern started his first week at Hone Capital, he heard me and my coworker chatting about Ethereum (and PepeCash, but that’s a different story). Not wanting to miss out on the “fun”, he opened a Coinbase account and bought some Ether. GDAX featured heavily in his browser’s history after that.

An artist’s depiction of Hone Capital’s office.

Naturally, when my second intern walked through the door and settled into her desk right beside Intern #1, she was fascinated by GDAX’s charts. She soon became an investor in Ethereum as well.

Hearing excited/panicked reports about Ethereum’s price on a daily (hell, near-hourly) basis, I began to notice a trend: Ethereum seemed up near the weekend and down near the middle of the week. Wednesday? ETH at $225. Saturday? ETH at $275. Was this actually a trend or was I making patterns from noise?

Hypothesis

I believe that sentiment/casual dabbling is the main driver of significant price movements for “mainstream” coins (applicable even before November’s cryptocurrency market cap boom). It’s not hard to see how — on Friday and Saturday nights — a buddy talks to his friend, a lady chats with her significant other, people gossip over dinner, and we end up with more casual money being pumped into Coinbase and all its listed currencies.

Alternatively, people hear crypto stories all week at work and then remember on Saturday to finally set up that Coinbase account.

Point is, my hypotheses were that:

1. Price increases of BTC, ETH, and LTC tend to be greater on Friday, Saturday, and Sunday (when casual interest gets spread and converted into action).
2. Price increases of BTC, ETH, and LTC tend to be smaller on Tuesday, Wednesday, and Thursday.
3. Price swings (in general) will be greater on Friday, Saturday, and Sunday (sentiment being magnified by casual investors).

Findings

After crunching the numbers (nothing too complex; the methodology is in the Data below), it turns out I was mostly wrong on all three hypotheses.

Chart 1 (just below) shows the median price change per day of the week. Sunday, at least at the median, has the worst drops across all three currencies on Coinbase. Conversely, the end of Sunday is the best time to buy. On the flip side, prices consistently rise between Sunday and Monday. Maybe people get excited about cryptocurrency during the weekend and use their Monday procrastination to buy some coins?

Chart 1. Green is relatively high, red is relatively low. Treat all rows separately.

At least my hypotheses partly survived the experiment: Saturday is a usually good day for cryptocurrency prices, while Wednesday tends to be a bad day (unless you’re a Litecoin enthusiast).

Median price movements only tell part of the story though. Here’s Chart 2, the average price change per day of the week. Try overlaying this chart with Chart 1.

Chart 2. Obvious conclusion when comparing with Chart 1 is that there’s a lot of confounding outliers. Who would have thought?

When factoring in outliers, which would be captured in an average, the weekend looks really bad. One interpretation of this data is that casual trading still happens on Friday, Saturday, and Sunday, but affects downward price pressure disproportionately. In other words, casual traders freak out about BTC, ETH, and LTC much more than they get irrationally hyped up about those. If you’re looking to flip weekly, the trend still holds: buy at the end of Sunday. Sell sometime before the end of Wednesday.

That guideline holds at the median and at the average, but we all know cryptocurrencies have too much daily volatility to consider just those data points. Chart 3 is a set of three box plots (minus the whiskers, which distort the appearance of the graphs dramatically) that highlight the interquartile range of daily price changes for each cryptocurrency.

Chart 3. Each bar represents the range between the 25th percentile and the 75th percentile, with the middle line illustrating the median daily price change.

If you are looking to create a day-dependent trading strategy for a particular cryptocurrency, you’d probably start with these charts. From visual inspection, BTC and LTC tend to exhibit the same behaviors, with the notable exception of Thursday and Friday being flipped. ETH, however, has its own set of tendencies. Many reasons could explain ETH’s deviance from the norm established by the other two coins, including:

• Differences in demographic mix, including age, occupation, investment size, and nationality
• Impact of mainstream endorsements in amplifying or dampening opinion among the demographic mix (and when those endorsements are typically announced)
• Magnitude of market manipulation by whales or institutions

These are all factors that could feed into a cryptocurrency weekly trading strategy. I’ve never tested the efficacy of bots or other programmatic methods to extract sentiment/price signal from public data, but such tools could be deployed according to the factors listed above.

If you came to the conclusion that BTC and LTC were highly correlated (a widely-held assumption that I’d like to add nuance to in a future Medium post) from these charts, no one would blame you. However, these charts are aggregations of hundreds of price movements: a third-quartile price increase in LTC on Monday one week might be matched by a third-quartile price increase in BTC on Monday the next week, rather than occurring on the same exact day. Actionable advice? Treat trading decisions (at least those decisions that are based off weekly cyclicality) of each cryptocurrency separately.

Another method of representing the magnitude of price swings is through standard deviation, which I’ve included in Chart 4.

Chart 4. A numeric depiction of the crypto rollercoaster.

My third hypothesis was also off the mark: Sunday is actually the most “boring” day of the week across all three coins. You’re most likely to be on the crypto rollercoaster on Thursday, though there’s still decent odds of a wild ride on Wednesday and Friday.

Between these various statistical measures, I think the building blocks of a sentiment-inspired trading strategy are available. You might get more mileage through the use of other forms of sentiment (e.g. Google search trends), rather than pure technical analysis, but this will hopefully suffice as a start. I’m certain that you will come to your own conclusions from all the charts and numbers above, but if you wanted to gather the data and do additional manipulations, my methodology is below.

Data

Fortunately, daily data on BTC, ETH, and LTC are publicly available and easily accessible. With a simple copy-paste of historical data from CoinMarketCap into Excel, I had all the data I needed.

This feeds straight into Excel as a nicely formatted table with no clean-up necessary.

To compare price changes, I calculated daily percent change in the close price during the date range of January 1, 2017 — December 20,2017 (year-to-date). Possibilities to extend this type of analysis:

• Investigate the bottom and top deciles (i.e. the top and bottom 10% of percent changes in price).
• Measure the absolute value of percent changes in price, considering just the magnitude of movement.
• Add nuance to correlations between cryptocurrencies: for example, are large movements in prices correlated differently than small movements in prices?

I’d love to hear your thoughts on cryptocurrencies! Leave a comment below or send me an email at kendrick.g.kho@gmail.com. Occasionally, I post pithy blockchain/cryptocurrency musings on my Twitter as well.

In case the SEC is reading this: the data provided in this post is for information purposes only and should not be construed as investment or tax advice, nor as a recommendation to buy, sell, or hold any particular coin/token/security.