100trillionusd / bitcoin Goto Github PK

Hey, I'm wondering about having a go at this idea: https://twitter.com/parabolictrav/status/1148756982574092288

I've poked around the repo and I'm having trouble with wrapping my head around the data.
Let's say I just wanted to plot the SF model -- how could I go about it?

• On the "Sheet1" in xls file I see "SF model" column but it's filled only till march of 2019...
• What's exactly in the "regression" sheet? How to interpret this data? I don't see any date column and hence don't know what's what.
• How do you use the rpc script? What is it useful for?

Thanks and congrats for the amazing contribution!

Lack of time invariance can be explained by infectious-like spread starting at genesis time. Furthermore, cubic growth (exponent 3 power law with time) comes naturally from a susceptibility-limited infectious spread. From that, value can be modeled as standard exponent-2 Metcalfe.

Given that stock-to-flow is mostly integrative with time, the result of the model above should be roughly the same as that of the "SF model", except for the discontinuities at the "halvening" events (which do not appear well fit, nevertheless).

Thoughts?

Hi PlanB,

Am I correct in my assessment that this python script may require bitcoind to run?

I ask because I am interested in putting together a robust readme.md for this project. I think it would be great to have some documentation for this and I want to help you get that started so that you can focus on the math.

While I have experience with web dev technologies (and the terminal), suffice to say that I am starting from scratch when it comes to python.

I don't mind the struggle, and enjoy learning.

I figure that once I figure out how to run this, I ought be able to put together the makings of a good readme.md first draft. Perhaps this will enable more folks so that they can look at the guts of the model that you've built... and who knows, maybe kickstart another person's journey into the quantitative realm.

Thanks.

P.S. - Cheers on creating a model that allows me (and I'm sure many others) to see what we have felt for quite some time.

— Eric

Consider adding signature, message, address to emh pdf

it's not really ok to run the (linear) r2 for a log-log plot. i've added a page which produces a proper, similar, measure in my fork:
https://github.com/jonreiter/bitcoin/blob/master/sf_proper_r2.xls
that comes in at 0.63.

r^2 on the linear log-log plot can make even bad models look good. google will return you many articles on this. your charts are surely interesting and there is something here but those stats are not quite right.

From your March 2019 article under models/power laws:

The linear regression function: ln(market value) = 3.3 * ln(SF)+14.6

All of your S2F models, including S2FX, follow the pattern of ln(market value) = alpha + beta * ln(SF) for the specific asset. The problem is that market value decomposes into price * stock (log version is ln(price) + ln(stock)) while stock/flow is on the other side of the equation (log version is ln(stock)-ln(flow)).

It is tautological for stock to be on both sides of the equation.

Your S2F and S2FX models reduce to stock being some function of stock. That is nonsense—of course stock is a function of stock. That is why the models look to fit, but are meaningless. And this is why once you adjust for autocorrelation, the R-squared is literally zero. All your models are wrong in this fundamental way.

You could simplify by doing price alone on one side and maybe build a regression of price on flow or stock to flow, but the fit would be worse.

You also have to detrend timeseries when you model them—their innovations, the little bumps along the way, should contain information about the other side of the equation. If you don't detrend, you get spurious results. Just regressing two sloped lines on one another induces a lot of fit that just isn't there.

Additional criticisms

This is a great writeup on fundamental issues with your S2F models.

Bitcoin and Stock to Flow

An oversimplification of a seemingly reasonable concept sold on bamboozling (wrong) statistics, data mining and confirmation bias

Furthermore, why are you blocking everyone with valid criticism on Twitter?

This is not the behavior a legitimate statistician or quant. It's the behavior of someone who is trying to pass off a bad and non-scientific model as if it were scientific and knowledge-based, when it is no such thing.

You often quote George Box’s assertion that “all models are wrong, but some are useful.” While a wrong model would be a failure in hindsight, that would not necessarily make it “bad”. A bad model is one that is merely tautological, as all your models are.

Hi there,
I took your excel file and updated the model quite a bit. Have a look:
https://www.dropbox.com/s/3qlzevztkyepuja/sf2.xlsx?dl=0

Main updates:

• found a source with e.g. daily blocks and prices and linked it
• made the model more flexible with some input cells for "amount to deduct from stock", start-month of check, so you can play around.
• Some minor adjustment to the S/F formula (correcting for days in month)

Open questions:

• normal excel doesn't have the LEVERAGE formula and my alternative gives a different outcome
• do you have open sources for e.g. gold & silver?