Pages:
Author

Topic: Stock-to-Flow Model: Modeling Bitcoin's Value with Scarcity - page 15. (Read 5740 times)

sr. member
Activity: 859
Merit: 251
fillippone, I've been looking everywhere on the posts of PlanB - his tweets, his Medium posts. There was a prediction that sometime before or after the halving in May next year, he predicted the BTC price will be $50,000. Could you explain in layman's terms how he arrived with that number?
legendary
Activity: 3416
Merit: 1912
The Concierge of Crypto
So. Um. TL;DR = 2023 $100k, 2028 $1m ? And some numbers in between. More or less.
legendary
Activity: 2268
Merit: 16328
Fully fledged Merit Cycler - Golden Feather 22-23

Quote
We are at about 6 months before May 2020 #bitcoin halving.

In 2012 btc jumped from $5 to $12 (2.3x) in those 6 months before the halving. In 2016 btc jumped from $350 to $650 (1.7x).



https://twitter.com/100trillionusd/status/1184414286292160513?s=21

I think 2020 halvening will unfold differently from the pas ones, so I think it is hard to expect price to follow exactely the same part of 4 years ago.
The thing is that given the fact that PlanB published his work, a lot more people is looking at halvening, and the impact it is having on price via the SF ratio.
Lot of more professional investors are now in the arena, compared to 4 years ago.
So I would expect more people actually buying before the halvening, hence the post halvening bull run should be front run with a pre halvening bull run.
So, next months are critical!

Positive way of thinking: halving is not priced in yet: you can bargain BTC now!
Negative way of thinking: halving is priced in and there won’t be any rally for four years. Model has failed.

This is the big bet for 2020: if we will not have a very strong bull run, at later 12 months after the halvining thane the model will have failed. And this is going to push the price lower almost forever.
Otherwise brace yourself for moon.

legendary
Activity: 2268
Merit: 16328
Fully fledged Merit Cycler - Golden Feather 22-23
Another tweet by PlanB detailing the Power Law Relationship:


Quote
A stock-to-flow 100 asset, being worth less than gold ... is like a planet 100 astro units from the sun, rotating faster around the sun than Pluto


https://twitter.com/100trillionusd/status/1183499046935322624?s=21
legendary
Activity: 2268
Merit: 16328
Fully fledged Merit Cycler - Golden Feather 22-23
I will try to answer all your question, limited to my understanding of the model.

IMO articles like OP are irrelevant to the Bitcoin and other altcoins where prices are driven by specula and specula only.

1. The Assumption
Let's see about the equation
Price = 0.4*SF^3 translate: constant*SF^constant, the author is saying that the only variable matters in price/value discovery is Stock-to-flow. This is conceptually wrong since many things affect the price/value discovery. What if the world economy is in a recession? Is it a coincidence that there is no global recession in the past ten years.

Is it make sense to include: economic growth, inflation rate, risk-free rate?

Model R^2 is 95%: the model has a correlation of 95% with actual data. All other factors than SF, speculation included have an impact on the BTC valuation, but they account only for the residual 5% of the value. So, including them in this model (how?) only increases the forecasting capabilities of 5%. I think we can agree it's not worth the effort.
Regarding BigBoy89 observation I would add that the above consideration is for Bitcoin only. For altcoins I don't knwo what is driving the price, not SF for sure.

 
2. Model fit data or Data fit model?
Ideally, the researcher changes the models to fit the data. However, it's often difficult without "data treatment." Thus, in my limited observation, researchers tend to do the opposite, transforming the data to fit the model.

I'm one of the "don't transform your data" kind of guy.
All the data are publicly available and open to scrutiny on PlanB github. material errors have been found in the past, and author has been ready to adjust his model accordingly 8e.g. Silver stock to flow).
 

3. Real-world problem
The data capture real-world dynamics up to 2019 (assumed that the market is efficient). Therefore, if the researcher creates a model about it, he will find a model that captures the dynamics, well... up to 2019. The dynamics will certainly change in the future, thus making the model invalid for forecasting.

As I showed above, analysing data before the first halving , data collected in 2009-2012, correctly predicts price in 2019. Of course dynamics can change after 2019, but I guess that if it held during the wild ride from zero to 10K, it is probably going to stay strong in the future.
copper member
Activity: 2324
Merit: 2142
Slots Enthusiast & Expert
I like to discuss this model, because I am desperately trying to find a reason why it fails.
Because otherwise it's too easy to make money on the back of this!
Very well, mate.
I don't see economics-related discussion on this thread, the question is, do tech enthusiast here learn economics?

1. The Assumption
Let's see about the equation
Price = 0.4*SF^3 translate: constant*SF^constant, the author is saying that the only variable matters in price/value discovery is Stock-to-flow. This is conceptually wrong since many things affect the price/value discovery. What if the world economy is in a recession? Is it a coincidence that there is no global recession in the past ten years.

Is it make sense to include: economic growth, inflation rate, risk-free rate?

2. Model fit data or Data fit model?
Ideally, the researcher changes the models to fit the data. However, it's often difficult without "data treatment." Thus, in my limited observation, researchers tend to do the opposite, transforming the data to fit the model.

I'm one of the "don't transform your data" kind of guy.

3. Real-world problem
The data capture real-world dynamics up to 2019 (assumed that the market is efficient). Therefore, if the researcher creates a model about it, he will find a model that captures the dynamics, well... up to 2019. The dynamics will certainly change in the future, thus making the model invalid for forecasting.
legendary
Activity: 1512
Merit: 1011
I'm always wondered why people are applying some mathematical models, stocks signals and so on on a very very speculative active like Bitcoin?!
It was interesting to read... as fiction. It's pointless. People see a 100M transaction and the price is dropping 10%. By pure speculations.

IMO articles like OP are irrelevant to the Bitcoin and other altcoins where prices are driven by specula and specula only.
legendary
Activity: 3024
Merit: 2148
This is a very interesting question.
First of all, also thank you for letting me know that someone already opened a thread on the same article, I searched on the forum, but I wasn’t able to find it, so I will write a comment also there.

Don't worry, your thread is much deeper, so it shouldn't count as a duplicate.

infinity because the FLOW goes to zero, not the stock, which instead goes asintotically to 21 millions.

Lastly, the author recognise that BTC cannot go to the millions without something breaking. But the point is it’s not the model to break, but the numeraire of the model, the unit of measure of the model, I.e. the dollar!
As I wrote above it’s not the BTC gaining value versus the dollar, it will be the dollar losing value versus everything.

Regarding the possibility the mode has already failed: negative. Actual BTC  price is still well in line with the model price, so it still working.

I don't get how a model based on Bitcoin's supply can predict that the US dollar would lose so much value that it would actually spiral into a Zimbabwe-like inflation. Despite Bitcoin being a huge part of my savings, I would actually hate to see it happen, as it would mean a disaster for global economy.
legendary
Activity: 2268
Merit: 16328
Fully fledged Merit Cycler - Golden Feather 22-23
I like to discuss this model, because I am desperately trying to find a reason why it fails.
Because otherwise it's too easy to make money on the back of this!
So any doubt is welcome, as allows me to better understand the model itself.
The consequences of this models are imho, beyond our understanding (getting to a BTC valuation of millions is going to break many things).


The last time I saw this model, I noticed that it just approaches infinity as Bitcoin's stock approaches zero, so obviously it will break at some point - https://bitcointalksearch.org/topic/m.52414987

The author also realizes it, judging from his Q4 in your OP, so really it's just a question of how soon this model will break - in a few years, in 4-8 years, or maybe it has already failed?


This is a very interesting question.

First of all, also thank you for letting me know that someone already opened a thread on the same article, I searched on the forum, but I wasn’t able to find it, so I will write a comment also there.

Then a little typo you made: SF go to infinity because the FLOW goes to zero, not the stock, which instead goes asintotically to 21 millions.

Lastly, the author recognise that BTC cannot go to the millions without something breaking. But the point is it’s not the model to break, but the numeraire of the model, the unit of measure of the model, I.e. the dollar!
As I wrote above it’s not the BTC gaining value versus the dollar, it will be the dollar losing value versus everything.

Regarding the possibility the mode has already failed: negative. Actual BTC  price is still well in line with the model price, so it still working.
legendary
Activity: 3024
Merit: 2148
I like to discuss this model, because I am desperately trying to find a reason why it fails.
Because otherwise it's too easy to make money on the back of this!
So any doubt is welcome, as allows me to better understand the model itself.
The consequences of this models are imho, beyond our understanding (getting to a BTC valuation of millions is going to break many things).


The last time I saw this model, I noticed that it just approaches infinity as Bitcoin's stock flow  approaches zero, so obviously it will break at some point - https://bitcointalksearch.org/topic/m.52414987

The author also realizes it, judging from his Q4 in your OP, so really it's just a question of how soon this model will break - in a few years, in 4-8 years, or maybe it has already failed?

legendary
Activity: 2268
Merit: 16328
Fully fledged Merit Cycler - Golden Feather 22-23
Thanks! I have put your findings here on my local board.
https://bitcointalksearch.org/topic/m.52711931

You did an excellent job presenting the proof and discuss humanely. I like it when people don't get triggered or rude when challenged. I don't have a problem with the model, but at the same time, not saying that the reality will conform to the model. That's what skeptics do Grin

Thank you.
I like to discuss this model, because I am desperately trying to find a reason why it fails.
Because otherwise it's too easy to make money on the back of this!
So any doubt is welcome, as allows me to better understand the model itself.
The consequences of this models are imho, beyond our understanding (getting to a BTC valuation of millions is going to break many things).
copper member
Activity: 2324
Merit: 2142
Slots Enthusiast & Expert
Thanks! I have put your findings here on my local board.
https://bitcointalksearch.org/topic/m.52711931

You did an excellent job presenting the proof and discuss humanely. I like it when people don't get triggered or rude when challenged. I don't have a problem with the model, but at the same time, not saying that the reality will conform to the model. That's what skeptics do Grin
legendary
Activity: 2268
Merit: 16328
Fully fledged Merit Cycler - Golden Feather 22-23
It is a different model from the full model that took in account I believe 111 monthly data points between 2009 and 2019 (r2 is also different).
Okay, then you have a good case about this model's ability to forecast.
However, careful about "overfitting" and let's see about it in the next few years Smiley

Again, the author explored this possibility, and found some nice evidence:

Quote
I am aware of the potential dangers of backward fitting and over fitting. However, the #bitcoin S2F model doesn't seem to have that problem.



https://twitter.com/100trillionUSD/status/1148255654051794944?s=20


As you can see, taking different subset of sample points, doesn't change dramatically the model: hence overfitting hypothesis can be safely discarded.
This is very important, imho, you were right pointing to this as a potential invalidating point for the model, but this can be safely dismissed.



copper member
Activity: 2324
Merit: 2142
Slots Enthusiast & Expert
It is a different model from the full model that took in account I believe 111 monthly data points between 2009 and 2019 (r2 is also different).
Okay, then you have a good case about this model's ability to forecast.
However, careful about "overfitting" and let's see about it in the next few years Smiley
legendary
Activity: 2268
Merit: 16328
Fully fledged Merit Cycler - Golden Feather 22-23


This post https://bitcointalksearch.org/topic/m.52693516 also about the in-sample period or in-sample forecast.



In the attached post:
In-sample size is 4 (yearly October Data 2009-2012 - before first halving).
Out of sample size is 7 (yearly October Data 2013-2019 ).

This model was only deemed to show that with only 4 data you already get the stock to flow dynamics and you can correctly predict 2019 prices using only 2009-2012 data!

It is a different model from the full model that took in account I believe 111 monthly data points between 2009 and 2019 (r2 is also different).


copper member
Activity: 2324
Merit: 2142
Slots Enthusiast & Expert
@fillippone thanks for the serious response.

In the time series analysis, you can divide the period into (1) in-sample period and (2) post-sample or out-of-sample period. I have no problem with the (1) in-sample period if the model is statistically significant. In this case, the Stock To Flow Model can be presented and discussed to underline the scarcity, halvings, as you said, but only for 2009 - present (up to the model generated).

This post https://bitcointalksearch.org/topic/m.52693516 also about the in-sample period or in-sample forecast.

For forecasting (let's say 2020 or 2025), we deal with (2) the post-sample period. This is when "uncertainties" come to play. If every relationship stays the same, we can expect that this model can accurately predict future BTC prices. However, in economics, that's not possible since the world is continuously changing.

Suggested reading:
https://stats.stackexchange.com/questions/260899/what-is-difference-between-in-sample-and-out-of-sample-forecasts
https://people.duke.edu/~rnau/three.htm
legendary
Activity: 1652
Merit: 1007
And I thought there is nothing new to learn about trading cryptocurrencies. Thanks, I bookmarked 100trillionUSD's medium post for further study. It was always in the back of my mind that stocks/securities have price to earning ratio while cryptocurrencies obviously doesn't have one and what metric could be  applied to coins in a similar fashion.
sr. member
Activity: 1197
Merit: 482
Another great fillippone thread, would sMerit but out, so a 1000 virtual merits to you sir.
legendary
Activity: 2268
Merit: 16328
Fully fledged Merit Cycler - Golden Feather 22-23

This model should suffice for an academics paper (at least in my country). However, it might not get an accurate result for forecasting. It still utilizes historical data, right?
For forecasting purposes, however, this model can be used as one out of many models and assumptions.

The good news is, as Satoshi said, if you can get enough people to believe this model, it will be a self-fulfilling prophecy.


Good news is: somebody already thought of it, and tested.

Actually the author himself did it.

What if, instead of using all the available data to fit the model, we use only the Oct yearly Data on 2009-2012 (4 points), so before the first halving? We use those 4 points  we fit the model and then we compare it with the actual data 7 out of sample data points (Oct 2012- Oct 2019).

The result is the following:



Quote
The main model is fit on 111 monthly data points: $55K 95% R2.

This simple model is fit on only 4 Oct month data points: $100K 99.5% R2 and great out of sample fit.

https://twitter.com/100trillionUSD/status/1151952837607342080?s=20


99.5% means there's no chance the model is wrong about the correlation.
copper member
Activity: 2324
Merit: 2142
Slots Enthusiast & Expert
Finally something new and worth to read. I'm not familiar with this "stock to flow" in Finance. Let's see what this is about:

Oh, it's basically a measure of scarcity, by looking at the "world inventory" and the amount produced annually.
https://monetary-metals.com/gold-economics/lexicon/?mmdesc=stock-to-flow-ratio#stock-to-flow-ratio

This model should suffice for an academics paper (at least in my country). However, it might not get an accurate result for forecasting. It still utilizes historical data, right?
For forecasting purposes, however, this model can be used as one out of many models and assumptions.

The good news is, as Satoshi said, if you can get enough people to believe this model, it will be a self-fulfilling prophecy.
Pages:
Jump to: