Weekly network forecast 25th February 2013

0.Introduction
Important: These posts are a test of the methods used to create the forecasts, and definitely do not forecast non-MTGOX US$ / BTC price related changes - for example a massive decrease in the cost of a hash ( something we expect any day now ). Please do not use the forecasts here for financial forecasts.

molecular reminded me obliquely that I was going to re-model Difficulty as a lagged function of price and Difficulty, so I finally completed it last night. The new model has a lower error than the current model, and was calculated using data up to the end of November last year, as for the other models. I've labelled the old method of forecasting Difficulty Method 1 and the new method Method 2. Model D1, D2 and D3 are the models used to forecast difficulty one, two and three retargets ahead.

Errors this week are significantly low, although this is probably due to the recent increase in hashrate from Avalon's and ASICMiners ASICs being added to the network.

1. Models and datasets:

The model datasets have been collected into one paste to save time. Model estimates have been likewise aggregated.

Forecast and canary model analysis

All datasets

All estimates

Canary model: An estimate of the current hashrate estimate based only on current price and previous network hashrates. Used to detect changes to the network caused by agents other than the direct action the MTGOX US$ / BTC price as determined by the model.

Models f1, f2, f3 and f4: These are the one, two, three and four week forecasts of the network hashrate, based on the modelled relationship between the MTGOX US$ / BTC and the network hashrate.

2. Results

• Canary model, Models f1, f2, f3 and f4: The network hashrate is greater than all models' 95% confidence interval lower bound. 
• One week forecast had -3% error.
• Two weeks forecast had a 5% error. 
• Three week forecast had 9% error.  
• Four week forecast had 5% error.  
• Canary model result had -3% error; 
• Difficulty modelling: The method 2 model has been acceptably accurate for the last two weeks, and is much better than method 1 at the longer range forecasts.
• One retarget forecast had 7% error.
• Two retargets forecast had a -3% error. 
• Three retargets forecast had -3% error.  

3. Tables and charts

Table 1:

One, two, three and four week forecasts of the weekly average network hashrate, as well as the canary model estimate.

Table 2:

A comparison of the four forecasts and the canary model estimate with the actual network forecast.

Table 3:

One, two and three retarget forecasts for the mining Difficulty at those retargets.

Tables 4 and 5:

A table of previous retarget date and difficulty estimates, and the actual retarget dates and difficulties as a comparison.

Chart 1: 

A graphical history of the one, two, three and four week forecasts of the weekly average network hashrate compared to the actual network weekly average hashrate.

Chart 2: 

A graphical history of the Canary model estimates of the weekly average network hashrate compared to the actual network weekly average hashrate.

Chart 3: 

A graphical history of the last twenty six weeks of retarget date and difficulty estimates, actual retarget dates and actual difficulties. It should be noted that each retarget will often be forecast by two consecutive weekly forecasts, hence the multiple points per retarget on the new chart.

Chart 4: 

A graphical history of the one, two, and three retarget forecasts of the Difficulty, modelled directly as a function of previous Difficulties and MTGOX US$ / BTC prices.

Donations help give me the time to analyse bitcoin mining related issues and post the results. 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

bitcoincharts.com: 1kqHKEYYC8CQPxyV53nCju4Lk2ufpQqA2

molecular: 1MoLECHau3nb9zV78uKraNdcZsJ2REwxuL