|
Title: My earlier google trend correlation still holding strong, crash predictable Post by: DamienBlack on July 04, 2011, 10:58:27 PM Earlier I started a tread (http://forum.bitcoin.org/index.php?topic=17318.0) about how price and bitcoin's google trend numbers are highly correlated. The only big price drop in the past was paired with a drop in the trend. I decided to see how well the current "crash" matches up with bitcoin's google trend numbers. It turns out, that the correlation still exists very strongly.
Image Below: http://img844.imageshack.us/img844/2894/yearu2.jpg The blue line is the google trend line for bitcoin, roughly scaled to match price. You can see that if we assume bitcoins popularity is the main factor driving price, that the current price is justified, and in fact, (I have to say it) lower prices might be seen. I know that correlation is not causation, but clearly there is a correlation here. It seems intuitive to me, since bitcoin is a lot like a commodity, that the number one factor driving prices is demand. Demand seems like it can be roughly estimated by getting an indication of interest, which is what google trends gives us. I think that assuming at least some causation is prudent. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: foggyb on July 04, 2011, 11:02:20 PM https://www.google.com/trends/viz?q=bitcoin&date=mtd&geo=all&graph=weekly_img&sort=0&sa=N
Sure looks flat to me. Just like the trading price. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: muyoso on July 04, 2011, 11:04:45 PM https://www.google.com/trends/viz?q=bitcoin&date=mtd&geo=all&graph=weekly_img&sort=0&sa=N Sure looks flat to me. Just like the trading price. You are looking at the last 30 days while the OP is looking at the last 12 months. Google does relative data scaling pertaining to the selected time frame. In the last 30 days they take the average of the data and set the average to 1. In the last 12 months they take the average over 12 months and set it to 1. Since nobody gave a shit about bitcoins until recently, the first 9 months of the 12 month graph bring the average way down. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: Mr. E on July 04, 2011, 11:07:32 PM https://www.google.com/trends/viz?q=bitcoin&date=mtd&geo=all&graph=weekly_img&sort=0&sa=N Sure looks flat to me. Just like the trading price. Looks like the peak was mid June (the 11th, 12th, 20th) and trending down since then. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: haydent on July 04, 2011, 11:11:10 PM OP your graph doesnt appear to actually predict anything (ie show something before it happens) it changes when the price changes. correlation but not prediction.
and the reason it is below near the end is that there is no price line to mach it against due to trading freeze, and then your graph ends. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: DamienBlack on July 04, 2011, 11:14:32 PM OP your graph doesnt appear to actually predict anything (ie show something before it happens) it changes when the price changes. correlation but not prediction. and the reason it is below near the end is that there is no price line to mach it against due to trading freeze, and then your graph ends. That is because google trends 12 month chart has a delay. Perhaps I should have used the term, justified, or expected, instead of predictable. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: DamienBlack on July 04, 2011, 11:16:48 PM Is there tabular data for that index? I would like to take a closer look at it and quantify that correlation a little more precisely. http://www.google.com/trends/viz?q=bitcoin&date=ytd&geo=all&graph=all_csv&sort=0&sa=N I'd love you to take a look at it, your statistical analysis is always very good. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: Bitcoin Swami on July 04, 2011, 11:17:08 PM OP your graph doesnt appear to actually predict anything (ie show something before it happens) it changes when the price changes. correlation but not prediction. and the reason it is below near the end is that there is no price line to mach it against due to trading freeze, and then your graph ends. I agree I dont think that crash at 32 had anything to do with lack of interest.. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: DamienBlack on July 04, 2011, 11:18:50 PM OP your graph doesnt appear to actually predict anything (ie show something before it happens) it changes when the price changes. correlation but not prediction. and the reason it is below near the end is that there is no price line to mach it against due to trading freeze, and then your graph ends. I agree I dont think that crash at 32 had anything to do with lack of interest.. When it was at 32, interest was at its peak. You can decide which way it relates. There is a case both ways. I think the price got to 32 because there was so much interest, but interest died down, and so did the price. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: Man From The Future on July 04, 2011, 11:19:38 PM I think it's more of a feedback loop than just search volume -> price.
Up price = news = speculators = up price = news, repeat, until speculators bail. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: DamienBlack on July 04, 2011, 11:21:25 PM I think it's more of a feedback loop than just search volume -> price. Up price = news = speculators = up price = news, repeat, until speculators bail. Why wouldn't the correlation break at that point? A huge drop in prices should generate news, which should lead to a higher trend then price. Or are you arguing that speculators haven't bailed yet? Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: Bitcoin Swami on July 04, 2011, 11:24:35 PM OP your graph doesnt appear to actually predict anything (ie show something before it happens) it changes when the price changes. correlation but not prediction. and the reason it is below near the end is that there is no price line to mach it against due to trading freeze, and then your graph ends. I agree I dont think that crash at 32 had anything to do with lack of interest.. When it was at 32, interest was at its peak. You can decide which way it relates. There is a case both ways. I think the price got to 32 because there was so much interest, but interest died down, and so did the price. Yeah who knows.. its a paradox marty. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: Man From The Future on July 04, 2011, 11:27:36 PM I think it's more of a feedback loop than just search volume -> price. Up price = news = speculators = up price = news, repeat, until speculators bail. Why wouldn't the correlation break at that point? A huge drop in prices should generate news, which should lead to a higher trend then price. Or are you arguing that speculators haven't bailed yet? (Above theory is probably wrong :)) Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: Man From The Future on July 04, 2011, 11:28:04 PM Is there tabular data for that index? I would like to take a closer look at it and quantify that correlation a little more precisely. http://www.google.com/trends/viz?q=bitcoin&graph=all_csv&sa=N I'd love you to take a look at it, your statistical analysis is always very good. All I got from that is a bunch of weird characters Code: ⁙⁵†⁵⁴††††⁴†⁸⁰⁴†⁴††⁇†⁔ How do you make Google give you the list? Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: imperi on July 04, 2011, 11:31:41 PM It works if you import it into Excel.
Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: KeyserSoze on July 05, 2011, 01:34:52 AM How do you make Google give you the list? You kidnap their children for ransom. Works every time. Just be sure to use a voice scrambler when you call or they will track you down. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: Hook^ on July 05, 2011, 03:22:25 AM I think it's more of a feedback loop than just search volume -> price. Up price = news = speculators = up price = news, repeat, until speculators bail. Why wouldn't the correlation break at that point? A huge drop in prices should generate news, which should lead to a higher trend then price. Or are you arguing that speculators haven't bailed yet? (Above theory is probably wrong :)) Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: StewartJ on July 05, 2011, 03:41:50 AM This is very interesting.
I wonder, have you considered if the Google Trend line should in fact be 3-5 days behind the exchange traded price of bitcoins? Once someone finds about bitcoins via Google, it does take quite a few days to get setup, funded and trading on those BTC exchanges. So there would be a delayed effect, assuming there is a causal effect between Google Trend and the exchange price of bitcoins. Just a thought. SJ Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: StewartJ on July 05, 2011, 04:05:59 AM This is very interesting. I wonder, have you considered if the Google Trend line should in fact be 3-5 days behind the exchange traded price of bitcoins? Once someone finds about bitcoins via Google, it does take quite a few days to get setup, funded and trading on those BTC exchanges. So there would be a delayed effect, assuming there is a causal effect between Google Trend and the exchange price of bitcoins. Just a thought. SJ Actually, I meant the reverse effect, regarding the delay... Google Trend line should be 3-5 days AHEAD of the exchange-traded price of bitcoins. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: DamienBlack on July 05, 2011, 04:10:46 AM This is very interesting. I wonder, have you considered if the Google Trend line should in fact be 3-5 days behind the exchange traded price of bitcoins? Once someone finds about bitcoins via Google, it does take quite a few days to get setup, funded and trading on those BTC exchanges. So there would be a delayed effect, assuming there is a causal effect between Google Trend and the exchange price of bitcoins. Just a thought. SJ Actually, I meant the reverse effect, regarding the delay... Google Trend line should be 3-5 days AHEAD of the exchange-traded price of bitcoins. It would be ahead/behind if the google trend line only represented the first time a user looked up bitcoins. In reality, someone thinking about investing in bitcoins is probably making lots of bitcoin related searches throughout the whole process. And people who own bitcoins probably search most of all. In the end, the "first search" that helps a user find out about bitcoins for the first time is probably only a small portion of the entire search volume. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: StewartJ on July 05, 2011, 04:24:45 AM This is very interesting. I wonder, have you considered if the Google Trend line should in fact be 3-5 days behind the exchange traded price of bitcoins? Once someone finds about bitcoins via Google, it does take quite a few days to get setup, funded and trading on those BTC exchanges. So there would be a delayed effect, assuming there is a causal effect between Google Trend and the exchange price of bitcoins. Just a thought. SJ Actually, I meant the reverse effect, regarding the delay... Google Trend line should be 3-5 days AHEAD of the exchange-traded price of bitcoins. It would be ahead/behind if the google trend line only represented the first time a user looked up bitcoins. In reality, someone thinking about investing in bitcoins is probably making lots of bitcoin related searches throughout the whole process. And people who own bitcoins probably search most of all. In the end, the "first search" that helps a user find out about bitcoins for the first time is probably only a small portion of the entire search volume. Good point. I find this intriguing, this possible correlation between google searches and the price of bitcoins. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: newMeat1 on July 05, 2011, 04:40:49 AM Quick, everybody go do a bunch of Google searches!!!!
Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: bitfreak! on July 05, 2011, 04:40:59 AM I was actually looking at the Google trend data for Bitcoin earlier today, and imo this trend definitely exists, there is a clear correlation, and it makes perfect sense, the OP explained it well. Bitcoin is a virtual commodity, one would expect supply and demand to play a significant role in the value of BTC. Google trends gives a very powerful way of measuring the growing popularity of Bitcoin. Such a drastic and sudden dip in the trend data was obviously caused by something other than chance imo, and that something was probably the Mt. Gox crash. A lot of miners got out, a lot of investors got out, and a lot of interest was washed way in the overall process. We can all thank Mt. Gox for this imo. I will never use their exchange, so help me god.
Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: StewartJ on July 05, 2011, 04:43:03 AM Quick, everybody go do a bunch of Google searches!!!! LOL !!! Let's make search bots !! Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: StewartJ on July 05, 2011, 04:48:20 AM Such a drastic and sudden dip in the trend data was obviously caused by something other than chance imo, and that something was probably the Mt. Gox crash. A lot of miners got out, a lot of investors got out, and a lot of interest was washed way in the overall process. We can all thank Mt. Gox for this imo. I will never use their exchange, so help me god. I cannot believe that the MT Gox incident will keep this bitcoin from thriving. If in fact we did scare the new crowd away, it's only temporary. I think we are at the final moments of getting our hands on cheap bitcoins, I have a strong hunch this is a great opportunity for us to be on the ground floor of it all. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: StewartJ on July 05, 2011, 04:51:57 AM https://i.imgur.com/Qxkcm.png This shows that Google Trends actually correlates pretty well with $/BTC. Some details: The graph you see on Google Trends is the 'relative scale' chart. But what I used for my analysis is from the 'fixed scale' trend to take out variations in Google Search traffic. The correlation coefficient between the two series is .86, which is not bad, but by delaying the Google Trend by one week that goes up to .96 which means that in general the trend tends to lead price by about a week. The graph is a log scale to make it easier to read, and the trend was normalized (divide by 3.51) and delayed to show how well it fits. And the price levels shown are a weekly average, since this is the resolution of the Google data. You might think that since the Google trend tends to lead price by about a week and it fits pretty well that you can use it to predict price. But, as you can see there is still quite a bit of variance. It can vary by as much as -100% to +137% with half the samples having a variance greater than 34%. (There is a zero term at the beginning of the series that does not show up on the graph because of the scaling) It is not exactly clear though that you can rely upon it as a predictive tool for projecting price, because that time shift in the later part of the sample does not actually produce a better correlation than the non-shifted series as a whole. My conclusion is that it is certainly an interesting correlation, and it bears watching. But the high level of variance, and inconclusive causal link to price make it a weak predictive tool at best. If you want to try and use it to trade use the fixed scale data, not the relative scale, divide the last term by 3.51 to get an approximation of price, place your orders, and be prepared to wait a week to see if they hit... But don't bet the farm. Well done Chodpod, you are the bitcoin data wiz! Wondering if this could me made into a live online chart somehow.... hmmmmm. SJ Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: bitfreak! on July 05, 2011, 04:54:34 AM This shows that Google Trends actually correlates pretty well with $/BTC. You sir, are a genius. VERY informative post. Can I use that graph else where? It's great.nt will keep this bitcoin from thriving. If in fact we did scare the new crowd away, it's only temporary. Yes, I agree with you, Bitcoin will come back strong given time, and the advantage is lower prices again for a little while, which I can't argue with, but it is a shame to lose the big influx of interest Bitcoin had garnered recently.Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: DamienBlack on July 05, 2011, 04:57:53 AM Thank you Chodpaba, an excellent analysis, as always. Very, very informative. I am actually surprised by the strength of the correlation, even though I pointed it out in the first place.
Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: StewartJ on July 05, 2011, 05:00:46 AM I am actually surprised by the strength of the correlation, even though I pointed it out in the first place. That was an inspired discovery. Well done! Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: StewartJ on July 05, 2011, 05:03:11 AM This would seem to be a decent indicator for longer term outlook in BTC investment, IMHO.
Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: DamienBlack on July 05, 2011, 05:05:38 AM For those interested, the most recent google trend numbers are:
July 1st - 36 July 2nd - 38 July 3rd - 38. Using the chodpaba method, that gives a price of $10.82. But as chodpaba pointed out, there is variance +- 100%, so don't go too crazy. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: StewartJ on July 05, 2011, 05:12:22 AM I'm sure it could be done. I once kluged a custom interface between an AS/400 and a Mac Pro in the 90's, but I'm probably not the guy. The tricky part would be to have the Google Trend data charted on a bitcoin price graph, 2 different data sources. I was thinking of someone like Cowboy who did that awesome bitcoinity chart. He uses his BTC donations for whiskey, btw. SJ Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: bitcoiners on July 05, 2011, 06:45:44 AM Here's another handy little Google tool for search history.
http://www.google.com/insights/search/#q=bitcoin&date=today%203-m&cmpt=q Shows relative search terms as well. The above map is for 90 days. You can change it however you like. Cheers. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: DamienBlack on July 05, 2011, 07:19:13 PM For those interested, the most recent google trend numbers are: July 1st - 36 July 2nd - 38 July 3rd - 38. Using the chodpaba method, that gives a price of $10.82. But as chodpaba pointed out, there is variance +- 100%, so don't go too crazy. Wow, that was quick. I guess we've reached google stability. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: enmaku on July 05, 2011, 08:21:10 PM I retract my previous conclusion that the Google trend is weakly predictive. It actually correlates best by lagging price data by about three days. You, sir, should be held up as an example in front of science classes. This is what it's all about folks: find a hypothesis that seems to match your data, test it vigorously until it seems proven to you, put it out there for the community to attack and defend it as best you can. The moment your hypothesis no longer matches your data, it must be abandoned or modified to fit the new data. I applaud you for your openness and bravery sir. You are a fine example for scientifically-minded individuals the world over. Now please teach my friends, family and coworkers how to do this. Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: DamienBlack on July 05, 2011, 08:24:47 PM @chodpaba
Wow, thanks for the higher resolution look. Very interesting. So it lags price by about three days, hmmm something to think about... Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: StewartJ on July 05, 2011, 08:34:57 PM @chodpaba Wow, thanks for the higher resolution look. Very interesting. So it lags price by about three days, hmmm something to think about... Well, it does over that time period anyways. I would not completely discount the possibility that it could be predictive, and would certainly pay attention to it if it diverges significantly. If it was AHEAD of price for 3 days it would make perfect sense, time for online discovery/investment of bitcoin. Do not understand the lagging... Title: Re: My earlier google trend correlation still holding strong, crash predictable Post by: DamienBlack on July 05, 2011, 08:41:01 PM @chodpaba Wow, thanks for the higher resolution look. Very interesting. So it lags price by about three days, hmmm something to think about... Well, it does over that time period anyways. I would not completely discount the possibility that it could be predictive, and would certainly pay attention to it if it diverges significantly. If it was AHEAD of price for 3 days it would make perfect sense, time for online discovery/investment of bitcoin. Do not understand the lagging... Perhaps higher prices tend to generate news and excitement. Imagine if you will Mr Average, he has about $2,000 in bitcoins. When prices go up, he is dreaming of becoming a millionaire googling bitcoin and reading every article on it, recommending it to all his friends who also google bitcoin. He is at best buy buying mining equipment, telling the employee helping him that it is for bitcoin mining. When the prices start going down, he hangs in for a few days, but quickly caves and sells most of his bitcoins. He doesn't recommend it to friends any more, he isn't buying equipment. Now that he isn't as invested, he isn't making bitcoin related searches as often. But, when prices rise again, he start googling to see if this time it is going to be sustained... and so on and so forth. |
