Yet another price prediction (This time for 1/1/14)

[altoidmintz]

The regression said 95% confidence  175 < $/BTC < 250 forecasted for 1/1/14. I said probably more like 200 < $/BTC < 240.
http://afterecon.com/economics-and-finance/historical-bitcoin-price-prediction/

Thoughts?

[1713551893]

1713551893

[1713551893]

1713551893

[1713551893]

1713551893

[1713551893]

1713551893

[adamstgBit]

just play with your math until you reach 500$ + for the next few months, k thanks.

Bitcoin is a new unknown element, tiny, but growing fast,  in totally messed up world of finance. the currency wars, the USD losing its reserve status, common poeple understanding the scam behind the fiat system

no, no amount of math can predict bitcoin price.

[mccorvic]

just play with your math until you reach 500$ + for the next few months, k thanks.

This is TA in a nutshell, be bull or bear

[nthoangga]

You squeeze to make a smaller range than the above predicted value. So actually your prediction and the artical's prediction are the same. Just you claim to have a more accurate value

[Melbustus]

You may think you have enough data to run a regression, but you don't. The underlying domain itself is changing constantly, so your sample is really a collection of tiny statistically-INsignificant effectively-unrelated samples.

This is the general problem with predicting any market's future performance based on past trade data.

[oda.krell]

First, note: I'm not slamming your method per se, altoidmintz. I for one appreciate any formal approach presented here. But you're not the first one to do regression analysis on $/btc, and if memory serves me right, they tend to undershoot.

But here's the easy/lazy/obvious reply to this kind of analysis: did you backtest it? In particular, did you apply the same method and ran the analysis on either the full data set, or a more recent subset, up to January 2013, and tried to look at the prediction for the following month(s)?

You can probably see where I'm getting with this -- unless I'm hugely mistaken, a sudden and violent rally like the one that brought us to the ATH in April almost certainly wouldn't have been within your 95% interval.

That said, if you're one of those who perform regression on the entire data set and then declare that's where price should be... well, in that case, we do TA from different perspectives then :D

[altoidmintz]

You may think you have enough data to run a regression, but you don't. The underlying domain itself is changing constantly, so your sample is really a collection of tiny statistically-INsignificant effectively-unrelated samples.

This is the general problem with predicting any market's future performance based on past trade data.

This criticism doesn't hold.

The regression is built using time series data which means that it accounts for normal fluctuations. I use well over 100 samples for the small model and over 1000 samples for the long term model which is statistically ideal. The model holds ceteris paribus. This is how real prediction is done.

It doesn't matter if the domain changes. It matters if the domain changes fundamentally and statistically significantly. Until some unusual event happens there is no reason to think that the model won't hold. The approach predictive modelling is to consider it an objective baseline to form real predictions from. If some pessimistic news goes out the numbers should be adjusted downward and vice versa.

If you are contending that the domain is changing now statistically significantly more than it usually does:
1) You need to back that hypothesis up. The null hypothesis would be that the statement is false.
2) If things are changing at unusual rates right now, as indicated in the article, that might signal a bubble rather than some inaccuracy in the model.

Also, look at the predictions and do a quick reality check. The numbers are very plausible.

[altoidmintz]

First, note: I'm not slamming your method per se, altoidmintz. I for one appreciate any formal approach presented here. But you're not the first one to do regression analysis on $/btc, and if memory serves me right, they tend to undershoot.

But here's the easy/lazy/obvious reply to this kind of analysis: did you backtest it? In particular, did you apply the same method and ran the analysis on either the full data set, or a more recent subset, up to January 2013, and tried to look at the prediction for the following month(s)?

You can probably see where I'm getting with this -- unless I'm hugely mistaken, a sudden and violent rally like the one that brought us to the ATH in April almost certainly wouldn't have been within your 95% interval.

That said, if you're one of those who perform regression on the entire data set and then declare that's where price should be... well, in that case, we do TA from different perspectives then

Nice points. Regression analysis is often conservative because many people making financial forecasts, myself included, would rather be safe than sorry.

That being said, it would be quite a bold statement if you were to claim that the price > $250 USD/BTC by January 1. That would constitute an all time high by more than 10% and would also require a change in trajectory because BTC has been on decline for about a week now. Certainly >$250 is possible, but possible and most likely are two very different things.

If you take a look at the data I did create two models, one for the long term and one for the short term, and backtest. The models yielded slightly different results. Interestingly, the coefficients for time was not statistically different which may indicate that the two models are roughly consistent. Regardless, I concluded that the short term model is more plausible due to the fact that it seems to constitute a weaker contextual extrapolation.

Thoughts?

[oda.krell]

Nice points. Regression analysis is often conservative because many people making financial forecasts, myself included, would rather be safe than sorry.

That being said, it would be quite a bold statement if you were to claim that the price > $250 USD/BTC by January 1. That would constitute an all time high by more than 10% and would also require a change in trajectory because BTC has been on decline for about a week now. Certainly >$250 is possible, but possible and most likely are two very different things.

If you take a look at the data I did create two models, one for the long term and one for the short term, and backtest. The models yielded slightly different results. Interestingly, the coefficients for time was not statistically different which may indicate that the two models are roughly consistent. Regardless, I concluded that the short term model is more plausible due to the fact that it seems to constitute a weaker contextual extrapolation.

Thoughts?

First, let's be clear: you extrapolate. With data set 1 (October 30, 2011 until October 30, 2013), extrapolating to January 1st (~2 months), you get 115 to 150 USD. In other words, we'd have to undergo another major correction. With data set 2 (June 1, 2013 until October 30, 2013), extrapolating to the same future date, you get to 175 to 250. Which would basically mean moderate growth or a mild correction in the coming two months.

And that's exactly the problem: if you use data set 1, you belong to the crowd of people that believe in the "long term trend" (i.e. ever since trading started). That's a plausible view, but the recent months simply don't support it. We have been, are and most likely will stay above that long term trend (which is basically what you regression on the entire data models).

If you use the recent data, I think you're getting closer, and personally, I find the numbers you get quite likely to be right. But my remark about "backtesting" went unanswered. What you meant by backtesting, I think, is the confidence with which you can predict the price on January 1st with your best model. What I meant was applying the best model to the truncated data, right before a major price move. It's pretty much impossible that the regression will capture the extreme price move, like for example the extreme rally in March/April. Some would declare such volatile phases "outliers" then, and ignore them, or consider them unimportant. I don't believe in that approach.

Anyway, I don't want to say that your analysis is useless -- it's just a big caveat one should keep in mind: if we continue to grow "as expected", your number will be right. If we will enter one of the (many!) extremely volatile phases, up or down, in the next 2 months, your numbers might well be off by a factor of 2 or more.