Weekly network forecast 3rd December 2012

0. Introduction
As promised in post 10.4 I've continued the long range forecast and canary models, to see just how well they manage to model the network hashrate into the future. I'll be using the original models, the details of which I've pasted:
Forecast and canary model analysis

Hashrates are weekly averages, MTGOX US$BTC prices are weekly medians.

1. Current datasets
The first group of pastes below are the data I used to generate the estimates. I've separated the models into separate dataframes so you can more easily reproduce my results and perhaps work on your own models, or update mine. Please note that the data is the log of the prices and hashrates, and the date is in unixtime.


Model 1 network hashrates, prices and lags
Model 2 network hashrates, prices and lags

Model 3 network hashrates, prices and lags

Model 4 network hashrates, prices and lags
Canary network hashrates, prices and lags

The second group are the estimate datasets with 95% confidence interval for the error. All data are in hashes per second, and the date is in unixtime.

Model 1 forecast

Model 2 forecast
Model 3 forecast

Model 4 forecast

Canary modelled estimate

2. Tables

 The first table below shows the network hashrate estimates created by the various models, along with the date at which they forecast to occur. More details on models f1, f2, f3, and f4 are in post 10.4; the details and reasons for the canary model are in post 10.3.

Simply put, the f models forecast the network hashrate at a particular date up to four weeks in advance. The canary model is based only on recent network and price history, and is more likely to detect significant changes from the model based only on previous network hashrates and the current MTGOX US$BTC price.

The "Error, Ghps" given below is the difference between the most recently completed estimate and the actual network hashrate at that date. Errors are assumed normally distributed and a 95% confidence interval calculated. More detail is provided by the error probability, which is calculated from the estimated normal CDF.

In todays error table, all error CDF probabilities are within a 95% confidence interval - between 0.025 and 0.975.

3. Charts
The charts below show the current and last twelve weeks of estimates, the forecast estimates in red and the actual hashrates in blue. The line range for each red point is the 95% confidence interval for that estimate.

The canary model shows the model estimates in red and the actual hashrates in blue, with the  line range for each red point is the 95% confidence interval for that estimate. When the estimate start to vary significantly from the actual network hashrate, it is possible that a force other than the MTGOX US$BTC price and previous network hashrate averages are affecting the network hashrate.

Donations help give me the time to analyse bitcoin mining related issues and write these posts. If you enjoy or find them helpful, please consider a small bitcoin donation:

12QxPHEuxDrs7mCyGSx1iVSozTwtquDB3r

Thanks to the following for use of their data:

blockexplorer.com:  1Cvvr8AsCfbbVQ2xoWiFD1Gb2VRbGsEf28

blockchain.info

molecular: 1MoLECHau3nb9zV78uKraNdcZsJ2REwxuL