Use Cases

[nehalbhanushali]

Prior work

• [x] Ron et. al [2] used the Union-Find algorithm to group accounts belonging to the same individual or entity. Analysis demonstrated that many attempts were made to obfuscate an entities’ transaction histories, finding many long chains and oddly shaped clusters in the graph that could not have been produced organically.
• [x] In addition, their research showed that while the net flow of the graph far exceeds the number of Bitcoins in circulation, most Bitcoins are in fact not in circulation and have not been moved or used since theirmining. This finding corroborates the hypothesis that significant money laundering for anonymization purposes is in effect.

[2] Dorit Ron, Adi Shamir Quantitative Analysis of the Full Bitcoin Transaction Graph, The Weizmann Institute of Science, 2012.

[nehalbhanushali]

• [x] Shah et al. claimed to produce a successful Bitcoin price prediction strategy based on Bayesian regression with a latent source model [3]. This model, originally developed by Shah to predict trending topics in Twitter, obtained significant success in predicting Bitcoin price, claiming a 50-day 89% ROI with a Sharpe ratio of 4.10, using 10-second historical price and Bitcoin limit order book features.

[nehalbhanushali]

• [x] Madan et al. used bitcoin blockchain network features, as well as seconds-level historical bitcoin price in historical time deltas of 30, 60 and 120 minutes to develop features for supervised learning. Leveraging random forests, SVM and binomial logistic regression classifiers, price deltas 10 minutes in the future were predicted, obtaining results of approximately 55% accuracy.

[nehalbhanushali]

MAJOR GOAL

The goal is to predict the price of Bitcoin in USD one hour in advance Also formulate as a classification problem based on whether price increases or decreases. (This is fundamentally a time series prediction problem)

Hence, All features must be computed using only information available an hour prior to the target prediction time

[nehalbhanushali]

• [x] In addition, we identified three addresses that had by the far the greatest influence on the network, with over 10% of all Bitcoin ever sent passing through at least one of these addresses. Creatively, we labeled these as A, B, and C, and of these the largest transactor (by over a factor of 2) was A.

ref [1]

[nehalbhanushali]

• [ ] Effect of twitter trends and wiki views on price. Only if there is already available data

[nehalbhanushali]

Previous results

Greaves et al. (26) analysed the Bitcoin Blockchain to predict the price of Bitcoin using SVM and ANN. The author reported price direction accuracy of 55 percent with a regular ANN.