Debunking the Lindy Effect for Cryptocurrencies
Antifragile, a book by Nassim Taleb, is a follow-up to the popular Black Swan that famously predicted the 2008 financial crash. It’s a favorite in tech circles, despite being composed largely of lengthy digressions and name-calling diatribes. Taleb’s thesis is that volatility is unavoidable and unpredictable; things that lose from it are fragile and things that gain from it are anti-fragile™️. Along these lines, one idea he introduces is the Lindy Effect, named after a New York Deli in which the following observation was made:
Broadway shows that lasted, say one hundred days, had a future life expectancy of a hundred more. For those that lasted two hundred days, two hundred more.
Generalizing, Taleb says that the life-span of non-perishables such as technologies can roughly be predicted by their existing age. That is, newer technologies are more fragile and have shorter expected lifespans compared to older, battle-tested technologies. Taleb points to bicycles and shoes as examples of technologies which have stood the test of time and therefore we expect them to continue to exist long into the future.
What does this have to do with Bitcoin? The recent explosion in crypto-currencies has led to people grasping for reasons to justify Bitcoin’s sky-high valuation. In this article by CryptoFundamental, the author invokes the Lindy effect to assert the anti-fragility of Bitcoin:
“Bitcoin is the only existing cryptoasset with any serious Lindy justification”.
The Lindy effect holds here because the cryptoasset market exhibits continuous extinction pressure on assets and protocols. The mere fact that a protocol or cryptocurrency has survived a meaningful amount of time in a competitive environment is an indication of value.
Again, another article asserts that Bitcoin will outlast other coins because:
The probability of long-term success [of Bitcoin] is [by] far the highest among cryptocurrencies, applying the logic of the Lindy Effect.
Yet another:
Bitcoin is the digital currency with the longest continuous use, and, as such, it is likely to be the digital currency that is continuously used in the future.
You get the point. These authors are all invoking the Lindy effect to elevate Bitcoin above other crypto-currencies. It’s true, Bitcoin has been around for a while; most other crypto-currencies are either derived from it or heavily inspired by it. However, it’s a bit silly to point to the Lindy effect to justify investing in any crypto-currency over another; following the implications of the Lindy Effect, an investor should only be investing in cash or gold and would be avoiding crypto-currencies altogether.
It also surprised me that none of these writers have actually fact-checked the Lindy effect. Wouldn’t it be a trivial matter to pick a point in time, find out the ages of the most popular coins at the time, and compare how well they have fared? I think the answer is that most people interested in crypto-currencies were too busy making boatloads of money to bother doing this research; luckily I am not so encumbered.
Hypothesis:
These authors are using the Lindy Effect to argue that Bitcoin is a better investment opportunity than other crypto-currencies. In other words, they are arguing that there is a strong correlation between existing age and future investment performance. So to test this theory, I picked a point in the past and evaluated how well a currency’s age at that time predicted the performance from then until now.
As a secondary question, I asked if there were other simple heuristics which work as well, if not better than the Lindy Effect. For example, I coin the “Snowball Effect” which is based on the idea that currently successful technologies will have the most growth. This would imply there is a strong correlation between current market cap and future investment performance. Testing the effectiveness of the Snowball Effect would give us a baseline to evaluate how powerful the Lindy Effect is.
Methodology
I used CoinMarketCap’s historical data to get a snapshot of the top crypto-currencies, by market cap, on January 05, 2014, and their Ticker API to get current data (as of 01/07/2018). After a little bit of data cleaning, I now knew how any investment made in January 2014 would have performed as of 4 years later.
Birth date of each crypto-currency was a little trickier, but I eventually found that most of their creations were announced by forum post¹.
Results
Let me preface this section by saying I am not a statistician. I used Google Sheets’ CORREL() function to compute all correlation coefficients, which may not be the right method for non-linear relationships. I tried a few different ways to measure growth, age and popularity, and measured correlations between them all. All data is in this Google Sheet, readers are welcome to run their own analyses!
Let’s start by looking at how well growth ratio was predicted. I defined growth ratio as:
growth ratio = (price in 2018)/(price in 2014)For example, Bitcoin went from $864.89 to $16135.80, a growth ratio of 18.66. On the other hand, Peercoin dropped from $7.21 to $6.25, for a growth ratio of 0.87.
Looking at the left column of the table, growth ratio was not strongly correlated with either age or age rank (oldest = rank 1, second-oldest = rank 2), neither having a coefficient greater than 0.1.
If instead we look at growth ratio vs. historical market cap and rank, we see that growth ratio vs. historical market cap rank had a weak but significant coefficient of -0.296. This suggests that the Snowball effect was at least as strong as the Lindy effect in predicting growth ratio.
I also looked at growth rank, such that the currency with the highest growth ratio was assigned rank 1, and so on. The Lindy effect was a better predictor of growth rank than of growth ratio, but the Snowball effect surpassed it by every measure.
Conclusion
It appears that for the time period from 2014–2018, the Lindy Effect has been a pretty bad heuristic for investing in crypto-currencies. This should not be much of a surprise to any crypto-enthusiasts, it’s widely agreed that their current prices are largely driven by speculation. Some might argue that the Lindy Effect will be more applicable when the bubble bursts, since the fragile coins should die when that happens. To this point, I’ll just point out that crypto-currencies were in a bubble in mid 2014, and several of them did “die”. I ran a separate analysis to evaluate the strength of that relationship, and didn’t see any effect there, but it was a bit tricky² to determine what constituted the lifespan of a crypto-currency.
So why did these authors try to use the Lindy Effect to justify investing in Bitcoin, without having any evidence of its validity? Did they have some evil plan to convince others to invest to increase their own profits?
It’s more likely they were trying to fool themselves. As Jonathan Haidt says in The Righteous Mind,
Anyone who values truth should stop worshipping reason…[Reasoning] evolved not to help us find truth but to help us engage in arguments, persuasion, and manipulation in the context of discussions with other people…[A logical reasoner] is really good at one thing: finding evidence to support the position he or she already holds, usually for intuitive reasons.
People in the tech community tend to think of themselves as logical people, and that the decisions they make must have been well reasoned. Unfortunately, this can lead to blind spots where we think we’re acting with cool, level-headed assessments, but are actually making choices with our emotions first and post-justifying them with reasonable arguments. If we think our conclusions are implied by facts, it can be even harder to change our minds. To avoid falling into this trap, we need to be cognizant of our own limited capability to perceive the truth. Ironically, I must agree with Taleb when he says, we need to “limit our downside” or risk being “fooled by randomness”.
Appendix:
1. Date of birth:
To determine each coin’s age at the time, for which I needed their birth dates. It turns out determining the birth date of each coin was not as obvious as I would have thought. Let’s look at some of the methods I tried:
Date of first entry in CoinMarketCap:
Would have preferred this, since I could scrape it programmatically. Sadly, after checking a few coins I realized this was largely dependent on when CoinMarketCap started collecting data. Litecoin, created in 2011, has no entries earlier than 2013.
Initial Github Commit:
This was another preferred approach, since I thought it could be easily automated. Unfortunately, this was made complicated by some coins not having Github repo’s. For the ones which did have repo’s, I thought it would be as simple as running:
git log --reverseBut it turns out that since many of the coins were forked from Bitcoin/Litecoin, this actually returned the date of the original commits. I did a little more digging, but was unable to find a way to find the first commit after a fork. (Maybe some git guru reading this can help?)
Whitepaper date:
Another promising approach, but many coins did not have white-papers (deservedly, since they only made small changes to fixed parameters of bitcoin). On top of that, white-paper date seemed to vary from date of public announcement quite often.
Announcement date:
It turns out the most effective approach was to Google search for “<coin_name> announcement”; most if not all cryptocurrency creations were announced on either bitcointalk or cryptocurrencytalk via a forum post by an excited user. Somewhat ironically, it turns out that these harbingers of decentralized consensus had their own history best preserved by centralized websites.
2. Attempting to measure the lifespan.
Before deciding to measure the investment performance of these crypto-currencies, I tried to measure the power of the Lindy Effect as written, which is a statement about the lifespan of a technology.
Death of a coin:
It’s surprisingly difficult to determine when a crypto-currency has died. Many crypto-currencies’ market caps dropped to effectively zero between July 2014 and June 2018, but rebounded in a huge way in July 2018. If this article had been written one year earlier, we might have considered those as dead. See Quark (QRK) below, which, had I conducted my experiment in 2015 or 2016, I would have declared dead.
Measuring health instead of death
Without a good way to determine if a coin has died or not, how can we evaluate the power of the Lindy Effect? Taleb’s thesis is that technology lifespans are largely determined by their level of fragility, which can in turn be predicted by the current age of the technology; surviving to that point shows some resistance to negative events. Instead of looking at date of death, we can look for any evidence that age has any statistical positive on the future health of a currency. As a proxy for health, we can use the relative rankings of Market Cap of each currency. If the Lindy Effect were powerful, you would expect to see some correlation between a crypto-currency’s age and it’s relative change in ranking.
With the list of top ~25 crypto-currencies from 01/05/2014, I used CoinMarketCap’s API to pull in each of their current market caps and did some basic data manipulation/cleaning. See this google sheet for all data extracted.
Result
Between age and relative ranking change, I found a measly correlation coefficient of just -0.079. Not only is this too small to indicate a real effect; it’s also slightly negative, the opposite of the effect we were looking for.
