I am putting together a report by overlaying recent Bitcoin price movements against the best matching past occurrences to see if there is any predicable continuation.
Currently there are no graphs or back testing to see how the different parameters affect accuracy.
I have some preliminary numbers based on comparing a dataset of 6 months, 8 months and 12 months then looking at the average outcome when looking forward from those matches out to four months.
The columns are weeks projected into the future and the numbers are percentages. So when comparing 6 months of data to find the most similar past pattern the best outcome after one week is a return of 115%, the second week is 117% etc. All of these numbers are based on the price data pattern as of July 10, 2018.
A big disclaimer; this is not a prediction, just because the past data set closely matches does not guarantee that the trend will continue.
These are just a small sample of results with random parameters. When using a larger pattern to compare with past data can result in more specific matches but at some point reduce the amount of range to search through.
A big issue with this approach when comparing patterns on historical data is tuning the parameters to the dataset rather than fundamental patterns. This comes through in the fact that the 6 month pattern predicts a decline in price and the 8 and 12 month data set predict an increase in price.
I will have to experiment with back testing different parameters to get a good sense of accuracy if any on historical data and the future prediction. That will come later.
Update July 11, 2018
Here are a small sample of back testing results. The data samples are collected in days back from 180 all the way back to 1275 three and half years ago. The data sets were run using 6, 8 and 12 months of data respectively and the results are shown as projected average divided by the actual percentage change ninety days later.
|6 Month Set||89/85||105/100||102/177||130/156||120/176||188/199||260/48|
|8 Month Set||46/85||93/100||121/177||110/156||116/176||216/199||199/48|
|12 Month Set||214/85||113/100||123/177||129/156||148/176||265/199||141/48|
Overall the results were kind of interesting. There are some correlation but also some significant misses. More resolution and adjusted parameters may make this a more useful report.