As I grow older I find myself wishing I'd taken mathematics more seriously so about the most I can add to this thread is THANKS for inspiring me to learn something. When I glanced at this ten second animation on youtube: http://www.youtube.com/watch?v=1hGsKphwC-A - demonstrating Linear Regression I realize I've put it into practice locating median lines on charts all along.
I noted that when Bitcoin broke out over the previous all time high, 266 or so, that the altcoin market lit on fire. I believe it's safe to say that all most all the altcoins experienced 10x to 20x growth in a very short time. Now when the new all time high was reached and price finally started to retreat, I noticed that the declines of the altcoins were much more extreme percentage wise than Bitcoin.
It seems to me that during this early phase, the past two months specifically, that one could seek safety in Bitcoin on downtrends and seek high beta in altcoins on the way up. I believe a lot of traders see this and eventually the trade is too crowded (for hobby traders to win at) but it certainly caused some excitement over the past 60 days.
Topic: Cross Sectional Regression Analysis of 35 Altcoins (Read 1607 times)
Well it can't predict anything- what it can do is describe the price given the dependent variables at a certain point in time (cross-sectional).
So right now based on the data Dogecoin is like $0.0000001 but that doesn't mean it will always be. In fact if it predicts anything, it predicts that ALL alt coins will appreciate over time *until something fundamental changes and then the model ceases to work.
So right now based on the data Dogecoin is like $0.0000001 but that doesn't mean it will always be. In fact if it predicts anything, it predicts that ALL alt coins will appreciate over time *until something fundamental changes and then the model ceases to work.
Kindly explain for the non-econometric savvy a little more about the model, how it works, how you construct it and what not. If possible...
Dogecoin is virtually worthless according the model.
Quote
Sorry to be so ignorant, but how does this let you make money? I'm just guessing that this might enable price spread prediction? And you could make money trading when market prices deviate from what your model says it should be?
Dunno.
I guess I should read the econometrics article on Wikipedia.
Dunno.
I guess I should read the econometrics article on Wikipedia.
From what I understand he is planning to make another altcoin with optimal parameters. However if his model really has r^2 of 0.85 he would probably be better off buying undervalued new altcoins.
As a side note you mind telling us what your model predicts dogecoin to be worth? I have a feeling that your model does not account for the variables that made some coins like dogecoin popular.
Sorry to be so ignorant, but how does this let you make money? I'm just guessing that this might enable price spread prediction? And you could make money trading when market prices deviate from what your model says it should be?
Dunno.
I guess I should read the econometrics article on Wikipedia.
Dunno.
I guess I should read the econometrics article on Wikipedia.
I didn't fit anything by hand, only ran an ordinary least-squares regression. There is data outside the model -- because there are at least 20 more altcoins with such low interest that I didn't include them but that have data -- and they predict their current valuations quite nicely.
There isn't any machine learning algo going on here just vanilla econometrics.
There isn't any machine learning algo going on here just vanilla econometrics.
Machine learning expert here. What it seems like you have done is over fit your parameters to your training examples. Have you tested your model with data points outside of your training set?
I would suggest that you add a normalized parameter to your regression cost function to prevent overfitting the training set.
Once you have done this please update your results, assuming you are not trolling.
I would suggest that you add a normalized parameter to your regression cost function to prevent overfitting the training set.
Once you have done this please update your results, assuming you are not trolling.
Well, now the number of altcoins out there is approaching 100, and doing a time-series analysis on what influences the dollar price of Bitcoins ran into a quagmire of auto-correlation that rendered the model not as useful as one might hope.
But now I ran a cross-sectional multivariate regression of 30+ altcoins using a number of variables with an adjusted R^2 of about 85%. Not bad!
So now the only thing left to do is use this info to create my own valuable new alt coin! mwuhahahaahaha!
For now I have obscured the variables because I plan to make myself some $ with this 1st
But now I ran a cross-sectional multivariate regression of 30+ altcoins using a number of variables with an adjusted R^2 of about 85%. Not bad!
So now the only thing left to do is use this info to create my own valuable new alt coin! mwuhahahaahaha!
For now I have obscured the variables because I plan to make myself some $ with this 1st
Jump to: