Comment by ionasiognio | Hacker News Reader

Comment by ionasiognio | Hacker News Reader

Parent Full thread

I see a few problems:

1. The author claims the lower bound is above 0 because "Bitcoin offers people a money storage and transfer system with two key properties: (i) permissionless access and (ii) decentralized database management." That might be true if Bitcoin were the only technology that offered those features, but it isn't. Cash is the most notable competitor, but there are also dozens of other cryptocurrencies which are drop-in replacements for Bitcoin.

2. The number of Bitcoin users has a hard cap because the technology is not scalable. It is just plain impossible for a large number of people to use it. This is not like Facebook, where the system is scalable and user growth is determined almost entirely by user interest.

Also, those graphs make me nervous. Fitting a line to a log-(log-square-root) graph using linear regression sounds awfully fishy to me. Would anyone with a better grasp of the statistics care to chime in?

> Fitting a line to a log-(log-square-root) graph using linear regression sounds awfully fishy to me. Would anyone with a better grasp of the statistics care to chime in?

It's fishy, but depressingly common in Economics. There's an old joke that 'everything on a log-log plot is linear when drawn with a fat magic marker'.

The errors at the top-right of the chart are much larger than the errors on the bottom left: they just don't look that way because of the log scale. They're also systematically biased at the top-left.

If you want to run OLS on a timeseries, you need ergodicity (i.e. the time series forgets its own history) for the regression to be valid. The number of observations also needs to be across a time much larger than the relaxation time of the timeseries. I don't see that here.

The entire thing triggers my spider sense for cranks. It's written in word and uses excel for its plots, the author runs some sort of crypto fund, and the fact that it uses 'square root of time' as a regressor is just bizzarre.

I'm no expert but here's a paper on using least-squares to fit a Gompertz curve: https://www.sciencedirect.com/science/article/pii/S037704270...

  but there are also dozens of other cryptocurrencies which are drop-in replacements for Bitcoin.

I don't see any drop-in replacements. I can see your point if you meant sharing a codebase. However, if you look at it as if it were a business, brand & marketing count for a lot.

Reply on news.ycombinator.com