Building prediction markets on Ethereum
This is the first in a series of posts on using quantitative models to help inform investment decisions in the crypto asset class, a highly undeveloped field of study
Crypto assets are a nascent asset class, with very little knowledge among investors on how to best analyze, value, and trade them. The short history of the asset class also means there’s very little data available to train models on. This partly accounts for why the majority of crypto funds today are discretionary, and typically have VC-like investment strategies. This is in contrast to traditional markets, where four of the five biggest hedge funds rely purely on quantitative systems to decide how to deploy capital.
Quantitative funds look to achieve returns through the systematic exploitation of market anomalies. Market anomalies are repetitive patterns in the market in which assets with certain characteristics consistently generate above-average returns. These anomalies exist because of certain structural reasons in the market (e.g. such as domestic capital controls) or because of repetitive behavioural patterns of market participants (e.g. such as herding).
Human behaviour is relatively consistent, meaning there’s strong reasons to believe that traditional market anomalies will exist in the crypto asset space as well. Given the lack of sophisticated quant funds arbing these anomalies, the returns of trading these anomalies should theoretically be much higher than traditional markets.
Introducing the Momentum Anomaly
The momentum anomaly is one of the most widely known (and simplest) anomalies in the market. It can be boiled down to:
On average, assets that have outperformed in the past will continue to outperform in the future, and vice versa.
From a high level, this essentially means investors should look to buy assets whose prices have gone up in the past, and sell assets whose prices have gone down. While this may seem counterintuitive to many, it’s important to understand that there are solid reasons for its existence and that the anomaly mainly exists on a short-term time horizon.
Theoretical Underpinnings of Momentum
The momentum anomaly has been documented in almost every single asset class. While there are likely a number of different reasons, I propose three fundamental factors that explain it’s existence:
Reason #1: Information is diffused and acted on slowly
New information travels with a time lag and is acted on with a time lag (after all, humans take time to process information). Thus, news may lead to sustained movements in one direction in a stock rather than a sudden spike, as more market participants become aware of it and act on it. Investors have different timeframes for their investments, and thus act on new information at varying speeds.
Reason #2: Liquidity constraints
Large investors are often limited by the liquidity of an asset, and thus must spread their purchases over a long period of time to avoid paying the spread. This results in long periods of buying, which manifests itself in sustained price trends, rather than sudden price jumps. This is also why price momentum strategies are typically more effective in smaller relatively illiquid assets.
Reason #3: Investor herding and return extrapolation
Fear of missing out is a well documented pattern of human investor behaviour. When assets have strong returns, it can lead to a herd mentality among market participants. This is amplified by the fact that human investors tend to extrapolate from the recent past when predicting the future, meaning high recent returns lead to optimistic forecasts of the future. Because of this, price trends tend to become a sort of self-fulfilling prophecy in which buying begets high returns and therefore further buying (at least in the short-term).
Testing Crypto Momentum Empirically
While momentum is a relatively simple idea, there are a number of different ways to slice it. We’ll start with the most basic type of momentum, called time-series momentum (TMOM). It was popularized in an academic study of commodity futures, where it was shown that TMOM had predictive value in all 58 assets tested.
It’s extremely simple: calculate the past return of an asset over a certain time-window, and compare the subsequent average returns of the asset when it’s TMOM is positive versus when it is negative. Theory states that the average subsequent returns of the asset should be higher when it’s in a positive TMOM state (i.e. current price is higher than prior price) versus when it is negative.
To test the effectiveness of the metric, we’ll calculate the TMOM of a universe of investable crypto assets. Our universe will be the five most well known crypto assets which have relatively long trading histories: Bitcoin, Litecoin, Ethereum, Ripple, and Dash.
For simplicity, we’ll use 10 days as our window for calculating TMOM and calculate it on a daily rolling basis. Thus, if the daily closing price of an asset is higher than the closing price 10 days prior, an asset is considered to have “positive TMOM”, and vice versa. We then compare the subsequent 10-day returns of the asset (calculated as the % difference in the closing price 10 days into the future from the current closing price). In reality, there needs to be much more stringent backtesting done to fully flesh out an anomaly, but we keep it simple here and just compare average returns for reader-friendliness.
Calculating this for the five assets in our universe for 2017, the basic summary results are pretty clear:
All five assets had higher average returns when TMOM was positive versus when it’s negative, although the results were only statistically significant for three. The most striking of which was Ripple (XRP), which averaged a 3.7% daily return when it’s past 10-day TMOM was positive (versus 0.6% when negative). These results hold after adjusting for volatility, with sharpe ratios all higher when TMOM is positive.
Looking at the previous two years (2015-2016), the pattern is the same, with the most significant result coming from ETH and LTC:
All five of the crypto assets had higher average returns following positive momentum states versus negative momentum states, which is interesting given that 2015 & 2016 were much different market environments than 2017. LTC and XRP actually had negative daily returns during this time period when TMOM was negative.
Going back even further to 2013 and 2014, we lose most of our universe except bitcoin and litecoin. However, the results still hold, with much higher average returns for both assets when they had positive momentum. Again, given how different the market environment was then, it’s interesting that this is the case.
The pattern holds when looking at other crypto assets as well. For example, here are the results for running this test for five more well-known crypto assets (NEM, Dogecoin, Monero, Vertcoin, & Bitshares) in 2017:
With the exception of Vertcoin, the average daily returns in positive TMOM states were significantly higher in the four other assets. In particular, NEM and Bitshares seemed to be driven primarily by momentum, with average return differentials of +2.5% a day between positive and negative TMOM states.
These results are repeated for this cohort when looking at the 2015-2016 years as well:
Again, all five of the assets had higher average subsequent returns following positive TMOM periods versus negative TMOM periods. In particular, the average returns of both Vertcoin and Bitshares were actually negative in the 10-day periods following negative TMOM periods.
While these backtests were overly simplified, the magnitude of return differentials between positive and negative TMOM states for some of these assets is striking. The consistency in the results is also surprising, with pretty much every asset tested having higher average daily returns when they had positive short-term momentum versus when they had negative momentum.
The message is clear, expect a crypto asset’s short-term returns to mirror it’s recent returns. In other words, in the short-term, it has historically been more profitable to buy into strength and sell into weakness.
Given the inefficiency of the crypto markets and the persistence of momentum in relatively established markets, momentum is something that serious market participants need to be paying attention. This is true even for long-term investors, as they can utilize momentum to better time entry and exit points, and improve their risk management.
Finally, given how simple time-series momentum is, it’s likely that more advanced versions of momentum will be even more predictive of asset prices. This will be the subject of the following reports.
Create your free account to unlock your custom reading experience.