By Being exposed to FUD in the Bitcoin space isn’t new. We’ve all experienced it at one time or another. While some FUDsters are blatantly wrong, some others hide behind complicated models that have the potential to trick a large number of people.
We’re going to look at why Digiconomist, an economist and data scientist, is so bad at modeling the Bitcoin and bitcoin mining spaces. Anyone who carefully looks at his methods would immediately brush his models to the side. After listening to Stephan Livera’s podcast with Ben Gagnon and Alex de Vries (Digiconomist), which you should too, I found five good reasons why no one should take these models seriously.
Digiconomist Proves He’s Bad at Modeling: A Drama in 5 Acts
It doesn’t take a rocket scientist to realize the issues associated with Digiconomist’s models, but it does take some thinking. Let me lay out for you in as simple a way as possible why you should ignore his models.
1. Modeling Highly Volatile Environments Using Fixed Variables
One factor of bitcoin mining that’s highly volatile is the profit margins. They can swing from 0% to 90% depending on many factors, including energy cost, bitcoin price, ASIC efficiency, government intervention, etc. So why would someone model this highly volatile factor with a fixed value?
Digiconomist tries to model Bitcoin mining profit margin by setting it at the fixed value of 40% because it’s the “long-term equilibrium.” That’s like modeling the temperature of your hometown at 50 degrees Fahrenheit while the yearly value can swing it between -12 F and 98 F. Anyone serious about modeling this value would at least use a simple regression model, not a fixed value. And according to him, “[t]he model stops working” when the market fluctuates away from the average. So why would he use a fixed-value model?
2. Over-Dependence on A Single Source of Data
When modeling complex systems, like the weather or equity markets, you want to take in as much relevant data from as many relevant data sources as possible. You might not use all that data as you discover it’s not necessarily useful, but you won’t rely too heavily on any single data set or source to tell you about the system you’re attempting to model.
This is what Digiconomist is doing. You’ll notice that throughout the debate on Stephan Livera’s podcast, he leans heavily on the Cambridge data. I have a couple of issues with this:
- Why doesn’t he reference any other data sources?
- Who’s to say that Cambridge’s data is any good?
Overreliance on any single data source, regardless of the fact that it comes from an alleged authoritative source, is bad.
3. Use of Old Data That’s No Longer Relevant
Models are made to attempt to predict the future. No model can predict the future, but they might get close. However, if the data you use to train, validate, and test your model is bad, your model might as well be a random number generator. Again, in such a volatile environment like Bitcoin and bitcoin mining, you want clean and recent data.
Throughout the debate, Digiconomist makes it clear that some of the data is from 2020. While I’ll give him the benefit of the doubt that not all of the data is likely that old, with these markets being so volatile any old data is unacceptable. At the very least, he should be trying to model that old data to estimate it to present values, but then you risk compounding error terms by embedding models within models.
4. Regime-Switching Models Are Lazy Models
Digiconomist said, “[o]ne model can be more correct in one situation” meaning sometimes the model breaks and you need another one. He then goes on to mention that he switches between three models when he believes the environment has changed states, with the states being growth, stability, and contraction. This is a big no-no because who’s to say which model should be used and when? You need something like a hidden Markov model to decide the current market state, not a human.
This is a regime-switching model and they tend to be terrible models if not used correctly, which is the case here. “[R]egime switching models are not suitable for capturing instability of dynamics because they assume a finite number of states and that the future is like the past.” Unless Digiconomist knows the total number of states the Bitcoin and bitcoin mining market can be in — which he can’t — he’s able to actually model those states properly — which, as we’ve seen, he can’t — and he knows exactly when each model should be used — which he can’t — only then can his models be trusted.
5. “The Models Can’t Be Verified”
When you build a model, you need to be able to verify that it has some level of accuracy and precision. That’s why using an isolated validation and testing process is so important so that you can analyze the efficacy and truth of your model.
According to Digiconomist, “the models can’t be verified.” I probably should’ve put this as reason #1 of why you shouldn’t take this guy seriously as it would’ve saved me some time writing and you some time reading.
Final Thoughts
It’s not a new thing to be using broken models to push a particular agenda and hoping that your audience can’t sniff out the falsities. We’ve seen this with Plan B’s attempted modeling of the Bitcoin price and we’ve seen it with Digiconomist’s attempted modeling of the bitcoin mining industry. The formulas and charts look pretty and when these “economists,” “quants,” and “data scientists” speak they sometimes sound smart, but they’re often either fooling themselves, fooling you, or both.
