Title: Random Walks and Logarithmic Growth (Fake Out). Post by: bitprick12345 on February 08, 2014, 08:45:46 AM http://imgur.com/UDVk82L (http://imgur.com/UDVk82L)
The problem with log scales is you can have a big drop but the fundamental floor is still ratcheting up. The regression based on the last two years shows the mid point is $500. The absolute floor looks like $350. But by 4.13.2014 the mid point will be $850 looks like its going to fake you out cause the water is rising. Its hard to know if Bitcoin will truly be a viral technology. But this is one way to explain what you see. In the short run bitcoin is over valued. But with log growth it does not take to long for the situation to reverse. Don't confuse this with a 50 day MACD either this is 2 years of data the predictive power of a long term regression has built almost every major investment firm you can think of. Title: Re: Random Walks and Logarithmic Growth (Fake Out). Post by: aminorex on February 09, 2014, 04:50:25 AM 4.13.2014 There is no 13th month in 2014.All kidding aside, if you want to be understood either use 2014-4-13 or 13 Apr 2014, both of which conform to international standards. Everything else is a local dialect. Wow, that chart's "prediction" really sucks. According to that chart you should never ever buy bitcoin, because it just goes up too fast. In fact, everyone should just sell it all now. Did Barry Silbert give you that chart? Title: Re: Random Walks and Logarithmic Growth (Fake Out). Post by: knightcoin on February 09, 2014, 10:04:23 AM If I was a day trader I would prefer put more attention into things like... search for "news sentiment" and "candlestick" correlations...
Once time I started write a little spider bot using the framework scrapy ( http://scrapy.org/ ) but because I am bad and lazy programmer I never actually finish it ::) :-[ ... but it was more or less like Get Sentiment Search ( such as ... bla bla .. check out http://www.makeuseof.com/tag/10-web-tools-sentiment-search-feel-pulse/ ) and Pin it into candlestick pattern triggers and then confirm it with western signal technical analysis :P Title: Re: Random Walks and Logarithmic Growth (Fake Out). Post by: bitprick12345 on February 09, 2014, 02:18:12 PM 4.13.2014 There is no 13th month in 2014.All kidding aside, if you want to be understood either use 2014-4-13 or 13 Apr 2014, both of which conform to international standards. Everything else is a local dialect. Wow, that chart's "prediction" really sucks. According to that chart you should never ever buy bitcoin, because it just goes up too fast. In fact, everyone should just sell it all now. Did Barry Silbert give you that chart? I guess thats one way to look at it. Its all about time scale. This is about a much longer term though process. In the states that date format is pretty common. As far as the prediction goes its doing pretty well. As the price has been converging towards the regression line. Title: Re: Random Walks and Logarithmic Growth (Fake Out). Post by: bitprick12345 on February 09, 2014, 02:35:31 PM I think what is being missed here is how you use this type of analysis. Most people are familiar with averages. In some way so much so that they forget what it is. It is a simple prediction metric, and always the best guess over any know data set.
If you take some points on a graph the average is the horizontal line that would be closest to all the other points. This is a valid prediction as it is a minimization of error. It is the absolute best number you can come up with if you constrain your analysis to a horizontal line. Linear regression allows us to slant that line to allow for growth or decay in the average if we are prediction prices etc. Log Regression gives us a curvature of that line to reflect accumulation of compound growth. This type of analysis since it is hard to over fit due to simplicity and is basically the limit of prediction. There is absolutely nothing that will out perform this type of model. |