So I'm practicing pulling data via php with the Havelock API and gretl. I pulled the entire transaction history of AM1 trades on Havelock, including days divs were paid and teased out day-of-week and month variables and ran a price regression weighted on volume. Dates ranged from 7/10/13 to the present. BTC-USD and BTC network data from Quandl.

*** denote degrees of statistical significance on t-normalization.

Model 1: WLS, using observations 79-8426 (n = 8348)

Dependent variable: Price

Variable used as weight: Quantity

Omitted because all values were zero: DMonth_6 DMonth_5 DMonth_4 DMonth_3

Omitted due to exact collinearity: DDayofWeek_7 DMonth_12

coefficient std. error t-ratio p-value

----------------------------------------------------------------

const -64.0891 7.37374 -8.692 4.27e-18 ***

Date -0.0200663 0.00281633 -7.125 1.13e-12 ***

Time of Day 3.51516e-06 2.03557e-06 1.727 0.0842 *

Last Div Amt 104.355 1.31024 79.65 0.0000 ***

Days Since FC Post -0.000763632 0.000133520 -5.719 1.11e-08 ***

BTCPrice24hAvg$USD -0.000432 2.06688e-05 -20.90 1.50e-94 ***

DailyVolume BTC$USD -4.32020e-07 2.54834e-08 -16.95 2.18e-63 ***

ln(BTCNetworkHashrate) -0.276515 0.0162034 -17.07 3.44e-64 ***

AvgBlocksize_MB_ 0.215881 0.0597392 3.614 0.0003 ***

Network Revenue 3.85365e-08 4.20636e-09 9.161 6.37e-20 ***

(daily, USD)

Total BTC in Circul 5.93937e-06 6.56003e-07 9.054 1.70e-19 ***

D_Sunday 0.00152049 0.00508091 0.2993 0.7648

D_Monday -0.0274101 0.00459092 -5.971 2.46e-09 ***

D_Tuesday -0.0385198 0.00470833 -8.181 3.23e-16 ***

D_Wednesday -0.0382756 0.00441648 -8.667 5.31e-18 ***

D_Thursday -0.0228972 0.00470844 -4.863 1.18e-06 ***

D_Friday 0.00268289 0.00446526 0.6008 0.5480

D_January 0.232385 0.00761263 30.53 6.22e-194 ***

D_February 0.235075 0.0119877 19.61 1.04e-83 ***

D_July 1.80466 0.0266744 67.65 0.0000 ***

D_August 1.33486 0.0209490 63.72 0.0000 ***

D_September 0.424720 0.0155508 27.31 2.90e-157 ***

D_October 0.0509887 0.0100955 5.051 4.50e-07 ***

D_November 0.0939666 0.00638384 14.72 2.00e-48 ***

R-squared 0.978181

Granted, the dataset is still very limited for a thin market such as Havelock, but a few things to point out:

- the variable "Days Since FCposts" is what is implies - the number of days since FC's last update to the forum. On average, since AM shares have traded on Havelock, each day that FC does not post produces a ~76k satoshi drop in share price. Granted, this is expected and the effect would become more drastic

- Total BTC in Circulation: this makes sense...as the total number of BTc rises, there are more available to buy shares, leading to an increase in the share price

*ceteris paribus *- Weekends are a good time to sell, it seems. Prices are higher, possibly because there are the most active users online/free at the same time over the weekend days (days' Correlation to measured relative to Saturday's contribution to predicted values).

- Since AM's capture of the high-water 30% network share, the price has decreased approximately 0.27 BTC per order of magnitude that the network hashrate has grown (that'd be your log effect)

Of course many of these variables are closely interrelated, but the Chinese exchange rate, the Chinese and American M0 supplies, and the LIBOR rate all produced negligible explanatory results when added to the model. The Chinese exchange rate would, of course, be pegged to the USD, so perhaps this is not entirely unexpected.

Feel free to PM/reply with suggestions/comments on improving this and other models. One useful variable I don't have is a time series of the shares outstanding on Havelock's exchange - this will definitely be useful in trying to explain, interpret, and rationalize buying and selling behavior over the past year.