BTC pricing models

[quantsig]

Hey -

I'm just posting some of our development models here to see if you guys might be interested in using any. Do any of you BTC traders use quant tools?

BTCUSD FX pair (gox)
http://blogdotquantsigdotnet.files.wordpress.com/2013/04/picture-15.png

BTCEUR FX pair (gox)
http://blogdotquantsigdotnet.files.wordpress.com/2013/04/picture-161.png

Gold/BTC ratio
http://blogdotquantsigdotnet.files.wordpress.com/2013/04/picture-171.png

All of the models are very young, so it will be some time before they mature. What do you think?

[camem]

don't worry this market's immature and volatile too - would be impressed if the same model worked for standard FX and BTC at the moment !

forgive my lack of quant knowledge, how much ahead of some historical data do you try and predict an 'output' ? A tick ? A fixed time period ? i.e. how far ahead is the model designed to look ? It looks from the close tracking of external events like it can't be much more than a day or two ?

[quantsig]

don't worry this market's immature and volatile too - would be impressed if the same model worked for standard FX and BTC at the moment !

forgive my lack of quant knowledge, how much ahead of some historical data do you try and predict an 'output' ? A tick ? A fixed time period ? i.e. how far ahead is the model designed to look ? It looks from the close tracking of external events like it can't be much more than a day or two ?

We have quite a few pricing models published at the moment. Many of them are traditional FX, but also Debt, Equity, and Commodities. All of which use our modeling engine. The currently published list is available here (chrome/firefox/opera): https://www.quantsig.net/#?&view=models

Your question is completely valid. There are different ways to approach the problem. The method you describe is quite common and has a proven record in various strategies. However, that is not the sort of models we run. Our models are trained to output the current price of the target instrument using a number of inputs from across markets. In effect, our models do not fit the future, they fit the market.

[camem]

ok i get it, it gives you a buy or sell snapshot

as far as training can't you use the three years of daily data we have historically ? or do you need more than that / consider the first couple not high enough quality ? would be interested to see more snapshots when you consider it trained

[quantsig]

ok i get it, it gives you a buy or sell snapshot

as far as training can't you use the three years of daily data we have historically ? or do you need more than that / consider the first couple not high enough quality ? would be interested to see more snapshots when you consider it trained

Output indicates what the model believes the instrument price should be. So a positive diverg can be considered potential undervalue vs. a negative diverg indicating potential overvalue.

Quality of data is a real problem going back in time with BTC. We decided to forgo attempting to massage historical data due to the low liquidity and poor quality of the available sources. As it is, we have been capturing live data properly for long enough to begin development, but not yet production models. Provided the modeler can handle the volatility (which we are very anxious to see), we should be able to publish production models within the next month or so.

[XXthetimeisnowXX]

ok i get it, it gives you a buy or sell snapshot

as far as training can't you use the three years of daily data we have historically ? or do you need more than that / consider the first couple not high enough quality ? would be interested to see more snapshots when you consider it trained

yes