10 Reasons that Explain the Score-to-Price Correlation Fallacy in Crypto Markets

Alice in Cryptoland: The Fool’s Errand of Trying to Predict the Market with a Single Metric

How many times have you heard an analyst of a crypto analytics vendor claim that they discovered a score function that is highly correlated with the price of Bitcoin or other crypto-assets? Some go as far as claiming that their magic-metric can serve as a predictor for price movements in a given crypto-asset. And, obviously, the data shown in the analysis is so compelling that you are ready to start trading based on the new magical-metric. Don’t! This is an example of a naïve belief in financial markets that single factors can serve as long term predictors of price movements. Many experts refer to that term as single-factor correlation fallacy. In the context of crypto-assets, I’d like to call this phenomenon the score-to-price correlation fallacy.

The temptation of trying to find a magic number that can predict the market is as old as the markets themselves. Even today, after 40 years of factor-investing theory, many financial experts and investors still try to actively beat the market following the trends of a single factor or metric. We even have single-factor ETFs! In large and rational financial markets, the single-factor trend has lost momentum in favor of more complex multi-factor strategies as most factors has proven to only be effective during specific market conditions in specific types of assets. However, a nascent and irrational market like crypto-assets brings back the hopes of discovering the magic metric that can solve the market. Ohhh well, the crypto markets are nothing if not fun 😉 However, I would like to outline a few ideas that might deter you from falling into the score-to-price correlation fallacy.

The idea behind the score-to-price correlation fallacy in crypto markets is fundamentally simple: crypto, like many other financial markets, is a complex and constantly changing environment that can be linearly predicted using a single metric. That explanation seems to make sense conceptually but it doesn’t offer any proof. Well, let’s take a look at several explanations that might force us to think deeper about this phenomenon.

10 Reasons that Explain the Score-to-Price Correlation Fallacy in Crypto Markets

Imagine that we are looking into a metric provided by a crypto analytic platform that is showing a strong correlation to the price of Bitcoin during recent time. The data certainly looks promising. Before you wire the funds, consider the following reasons about what could be wrong with your magical score:

1) Overfitting: Overfitting describes a phenomenon in machine learning and statistics in which a function(metric) is too closely fit to a set of training data points. In our example, the financial metric could be overly optimized for the Bitcoin price dataset used for testing showing remarkable correlations. However, the performance will quickly fall short when tested under regular market conditions.

2) Factor Contamination: Factor contamination describes a phenomenon in which a single factor or metric shows effective correlation with certain market conditions prompting its adoption as part of trading strategies. However, the adoption of the factor contributes to its decrease in effectiveness. Think about it, if your metric is used in trading strategies, then it becomes another attribute of the current Bitcoin market creating a recursive relationship with our metric.

3) Ignoring Outlier Events: As a programmable infrastructure, crypto markets are very susceptible to all sorts of outlier events: Block halving, forks, hacks, large blocks are some of the many examples that are nearly impossible to factor in linear metric models.

4) Different Factors will Contradict Themselves Under Different Market Conditions: Pick your favorite technical analysis platform and look at a dozen of technical indicators for a given crypto asset. You are likely to find that many of the different indicators will contradict themselves. While technical analysis can be certainly puzzling, this phenomenon is very common for any metric we choose. Our magic metric might perform well under specific conditions in the Bitcoin market but will fail under a different environment.

5) Markets are Non-Linear: Most, if not all, of the magic metrics we have seen in the crypto space are based on linear formulas. However, even though we don’t quite understand financial markets, it has long been established that they behave based on non-linear dynamics. From that perspective, predicting a non-linear Bitcoin market using a linear formula is a fool’s errand.

6) Ignoring Risk: There is no well-established definition of risk in crypto markets and, no real trading strategy is complete without evaluating risk. A metric that evaluates performance, and not risk, is a recipe for disaster.

7) Ignoring Macro-Factors: Like any other financial market, crypto is vulnerable to macro-factors such as momentum in equities market, regulatory policy, adoption by institutional and many others. At the moment, most single-factor prediction scores in the crypto space fail to incorporate macro-factors.

8) The Past is not a Reflection of the Future: In irrational markets such as crypto, trying to predict the future based on past behaviors has proven to be a frustrating experience. In a nascent market, many of the conditions have had no precedent in the past fooling any basic linear model.

9) The No-Free-Lunch Theorem: A classic idea in factor investing tells us that over a long-period of time, a portfolio of diverse factors tend to outperform the best of the individual factors. This is known as the No-Free-Lunch theorem and remains a strong argument against single-score models.

10) The Berkson’s Paradox: Berkson’s paradox is a famous phenomenon in statistics that describes a situation in which we look for association between two independent variables when there isn’t one. There are plenty examples of the Berkson paradox in the crypto space in the form of association of metrics like social sentiment or developer activity with the price of a crypto-asset.

These are some of my favorite arguments that illustrate the score-to-price correlation fallacy. Single metric models can certainly be effective under specific, short-term market conditions but they fail to act as long term predictors for an emerging and irrational asset class like crypto. Next time someone tries to sell you a single score as a predictor of the crypto market, please consider the aforementioned explanations and the….run!