Bitcoin Price Exploration (Part 1) Why Most People Shouldn’t Trade It

Part 1 - “Why Most People Shouldn’t Trade It.”
Part 2 - “Simulating What To Expect”
Part 3 - “What’s The Chance Of That Price Prediction Anyway”

This is part 1 of a series of articles where I will explore Bitcoin prices and returns in an attempt to learn something new and interesting. I am doing the analysis and writing the articles as I go, so it is very much a process of exploration.

The inspiration for part 1 came from a recent tweet by Matt Khoury where he shared an interesting chart from Fundstrat. The chart showed Bitcoin’s year by year return since 2013 and (this is the important bit) the effect on your returns each year if you missed the 10 best days.

The message was simple and powerful. If you try and time the market and you end up missing some or all of those 10 good days, your overall return will be substantially worse than a simple hodl (not a typo, hodl is a real thing in crypto). Are you smart enough to predict those 10 good days before hand and definitely get it right? I wish I could but I know I probably can’t.

Another way of looking at this is that most of Bitcoin’s mind boggling return has come from a very small number of fantastic days each year, and not from most of the rest of the time. That sounds to me like a power law distribution, something most humans find hard to deal with since our intuitions and evolutionary instincts are better suited to the linear and normal. Nassim Taleb calls this Extremistan.

For part 1 I decided to try and replicate Fundstrat’s results so that I could gain a better understanding of Bitcoins past price patterns, distibution of returns, and what this might mean for the future price. Part 1 will discuss the results of this, while future parts will dig deeper into related matters.

Quandl have a price history dataset from with Bitcoin prices going all the way back to around Aug 2010. I will be using an Ipython notebook to analyse the data. If you would like to see the code any try it out yourself there’s a Github link at the bottom of the article. First lets load the price data into iPython and then just to make sure everything looks ok draw a price chart.

Bitcoin price chart

That looks right to me. Notice the tiny bump towards the end of 2013. We’ll be coming back to that later. Out of interest let’s first find the 10 best and 10 worst days in the entire data set. Below are the 10 worst days. Quite a few days in the -20% to -25% range, and the very worst was a -64% day in 2010.

| Date | Return |
| 2010-09-15 | -64.62% |
| 2013-04-11 | -38.01% |
| 2013-04-10 | -37.67% |
| 2010-11-07 | -25.99% |
| 2010-10-11 | -23.90% |
| 2015-01-13 | -23.55% |
| 2011-10-17 | -22.87% |
| 2012-08-18 | -21.36% |
| 2010-10-24 | -21.05% |
| 2010-11-09 | -20.81% |

Now lets look at the 10 best days. The gains range from just under 40% to one staggering day of 173% in 2010. To be honest we probably need to take the 2010 figures with a pinch of salt. Those were early days for trading and exchanges. Even ignoring 2010 you can see how extreme the daily returns can be, both on the upside and the downside.

| Date | Return |
| 2010-10-07 | 37.93% |
| 2010-11-02 | 40.66% |
| 2011-05-12 | 42.01% |
| 2013-04-12 | 43.15% |
| 2011-04-29 | 53.70% |
| 2011-06-07 | 67.41% |
| 2010-11-06 | 72.41% |
| 2010-10-23 | 74.29% |
| 2011-01-31 | 90.00% |
| 2010-09-14 | 173.01% |

Next up lets group the data into years, drop 2011 and 2018 so that we are only left with full years, and chart the return by year.

If you thought 2017 was a good year, take a look at 2013. The price increased a staggering 5,408%. Remember that tiny bump in the Bitcoin price chart in late 2013. Well that huge 5,408% return in 2013 is now in the rear view mirror just a tiny bump, hardly noticable on a recent price chart. If the bull case for Bitcoin is correct, then when we look back in a few years time the huge increase and then drop at the end of 2017 / start 2018 will also look like a tiny bump. 2014 was the only down year, with a return of -58%.

Next lets find the top 10 days (by highest % return) for each of the years and see what happens if you exclude them. The chart below shows the year by year returns if you were fully invested (hodl’ing), and then compares that to the year by year returns if you missed the top 10 days each year. The tall blue bars are the returns from hodl’ing, and the short mostly negative orange bars the returns if you miss the 10 best days. Notice the large difference every year. If you missed the 10 best days each year your returns would not have been very good, in fact as we will see below you would actually have made a loss.

Bitcoin HODL versus missing top 10 days

Before we summarise let’s take a more detailed look at 2017 just to see an example for a specific year. Here are the 10 best days from 2017.

| Date | Return |
| 2017-07-20 | 27.96% |
| 2017-12-07 | 21.86% |
| 2017-11-13 | 14.58% |
| 2017-12-06 | 13.99% |
| 2017-09-15 | 13.69% |
| 2017-12-26 | 13.31% |
| 2017-12-11 | 12.72% |
| 2017-07-17 | 12.70% |
| 2017-08-05 | 11.97% |
| 2017-10-12 | 10.49% |

The compound return from those 10 days is 313%. If you missed those 10 days your return for 2017 reduces from a staggering 1,376% down to a “mere” 258%.

Finally to bring it all together let’s look at the compound annual return for an investor who hodl’d for the entire period (1st Jan 11 to 31st Dec 17), versus an investor who missed the top 10 days each year.

Compound annual return from 1st Jan 2011 to 31st Dec 2017
| | Return |
| HODL | 365.3% |
| Miss top 10 | -5.8% |
| Miss top 5 | 74.7% |

Think about that for a bit. If you hodl’d you made massive compound returns. But if you invested over the same time period but missed just the top 10 days each year you would have made a fairly bad loss (-5.8% per annum, or a loss of around -34% for the period). So all of those massive Bitcoin returns came from just 2.7% of the days (10 out of 365). The other 97.3% of the days made a negative contribution but you need them to make sure you catch the 10 fantastic days. Out of interest let’s also calculate the effect of missing the top 5 days each year. In that case your compound annual return is still very good at 74.7%, but significantly lower than the 365.3% you would have got from hodl’ing.

After writing the first version of this article, I decided to check and see how the same analysis would look when applied to the S&P500. Below is a chart of the year by year returns with and without the top 10 days, and a summary of the compound annual return for the same period.

S&P500 HODL versus missing top 10 days
Compound annual return from 1st Jan 2011 to 31st Dec 2017
| S&P500 | Return |
| HODL | 11.4% |
| Miss Top 10 | -8.3% |
| Miss Top 5 | -0.5% |

So it seems that Bitcoin is not special in this regard. If you miss the top 10 days on the S&P500 your compound returns are reduced from a respectable 11.4% per annum to a loss of -8.3% per annum. Miss the top 5 days and you still make a loss.

That’s the end of Part 1. In part 2 I will look more deeply into the actual distribution of daily returns and run some simulations to see what kind of price patterns can be expected in the future.

In closing here are a few of my thoughts on the implications of the above:

  • If you hodl but constantly check the day to day price, you will be miserable 98% of the time, and ecstatic 2% of the time. Not a good recipe for a happy life. Hodl’ing is really hard.
  • A better way to work out the percentage of miserable days hodl’ers suffer is based on analysing drawdowns over time.
  • The property that most of Bitcoin’s return comes from the top 10 days a year is not unique to Bitcoin. The same is true of the S&P500 even though the size of the S&P500 returns are much less extreme.
  • Contrary to what I have said so far, if you are a trader and know how to catch the 10 top days each year then please do get in touch (right now please).

If you enjoyed this article or got any benefit please clap so that others can find it too.

The Ipython notebook code is available on Github if you would like to try it out, double check my code and calculations, or hopefully add to and improve it.

(note - in the first version of this article the dates were all out by 1 day. The reason for this is that the Quandl dataset I used  contains the opening price for each date and the analysis assumed that this was the closing price. Annual calculations were also affected by this. The error has now been corrected.)
Nothing in this article should be taken as investment advice. If you got this far then you probably already realised that Bitcoin is extremely volatile. Before investing in Bitcoin or anything else please do your own research.