Lately I saw many novice traders make high-conviction market calls using the NVT ratio.
I was instinctively worried that they are overconfident about predictive power of the NVT, especially given that there isn’t any empirical evidence that it actually works.
So I downloaded historical daily data for Bitcoin from https://coinmetrics.io/data-downloads/ and did a quick and dirty test of statistical significance.
I use today’s NVT and look at whether or not it can predict the return between today and tomorrow.
(You may wonder why I look at 1-day return. It’s because if you look at a much longer time horizon, you effectively reduce the number of data points as the dependent variable would then be autocorrelated. Remember, Bitcoin doesn’t have that long of a history.)
So it turns out that NVT is a statistically significant predictor of 1-day return. (For people who have taken stat 101, the p-value here is 0.007). And the coefficient is negative, which makes sense: if NVT is high, we expect the market to fall.
But here is the catch. You are right only 55.5% of the time.
I am not surprised. Any individual signal will not give you a massive edge over the market. Not to mention the fact that the 55.5% is just an in-sample optimization.
So, unless you are an algorithmic trader who relies on the law of large numbers of to make a living, don’t expect NVT to make you rich.