Order flow and the bitcoin spot rate

Bitcoin is a decentralized, open-source cryptocurrency used to make private, peer-to-peer transactions anywhere across the world. Although the individuals involved are (mostly) anonymous, every Bitcoin transaction is a matter of public record; anyone can technically view every Bitcoin transaction ever made, including purchases and sales initiated on Bitcoin exchanges.  As such, the Bitcoin market lends itself nicely to the order flow methods developed by Evans and Lyons (JPE 2002). Following that methodology, this paper develops and estimates an order flow model of the Bitcoin exchange rate using a data set consisting of all major currency transactions occurring on the Mt. Gox Bitcoin exchange between June 26, 2001 and May 13, 2013. For each day individual transactions are separated into buyer-initiated (bid) transactions and seller-initiated (ask) transactions; the net of bid and ask transactions, order flow, in theory captures net demand in a market and should thus help explain and predict appreciations and depreciations in an asset's price. I specify the following estimating equation:

 Δs_t = λ₁z_t + λ₂i_t + ε_t,.

In the above expression, s denotes the natural logarithm of the Bitcoin spot rate, x order flow, i an overnight interest rate, and ε an error term. The overnight yield captures the opportunity cost of engaging in daily bitcoin arbitrage. Results are promising; R² values for the above expression estimated for U.S. Dollar-, Euro-, Pound-, and Yen-Bitcoin spot rates are approximately 0.60 and order flow is a significant determinant of Bitcoin spot rates.  Moreover, the models are generally robust to different specifications of the arbitrage yield.