Description

This repository accompanies the article https://doi.org/10.1016/j.frl.2020.101489 in which we analysed relationships between financial markets and the network of Bitcoin users (the so called Bitcoin users' graph). We were inspired by the idea of comparing the users' graph properties to Bitcoin price (by means of Principal Component Analysis). This idea goes back to Kondor et al., https://iopscience.iop.org/article/10.1088/1367-2630/16/12/125003/pdf.

The data used in the article https://doi.org/10.1016/j.frl.2020.101489 can be obtained using this repository. It is also possible to obtain the most recent data to keep the research up-to-date. The scripts contained in this repository can be used to obtain the time series derived from the Bitcoin users' graph. For the information about the time series:

• for the definition of the time series, see our article, Appendix, section 2. - the time series are denoted by s_i,
• how the time series defined above are incorporated into the main regression model, see our article, section 3.3 - the time series are defined there by the values UsersGraphProperty_i,j,t, where t=1,...,T and T is the number of subsequent periods taken into account.

The whole process of obtaining the aforementioned time series can be divided into the following steps:

1. Users' graph extraction - the creation of the graph is explained, e. g., in the subsection 4.1 of the Article of D. Maesa et. al., https://www.researchgate.net/publication/320026355_Data-driven_analysis_of_Bitcoin_properties_exploiting_the_users_graph. In the users' graph extraction process, we use the modified Bitcoin client of D. Kondor. The github repository of this client is available here: https://github.com/dkondor/bitcoin. Apart from the modified Bitcoin client, we use as well the following three repositories of D. Kondor: https://github.com/dkondfor/txedges, https://github.com/dkondor/join-utils, https://github.com/dkondor/sccs32s.

2. Long-term subgraph extraction - from the users' graph, we can extract as well a subgraph (called the "long-term-subgraph") which contains users who were active in a period long enough, took part in sufficient amount of transactions, and are represented by sufficient number of Bitcoin addresses. The idea of looking at the long-term-subgraph goes back to D. Kondor et. al, https://iopscience.iop.org/article/10.1088/1367-2630/16/12/125003. This concept has been also used in our article (see Appendix, 2. Obtaining values of variables from the users graph).

3. Snapshots' creation - we create snapshots for the periods taken into account as described in our article in Appendix in section 2. The snapshots are denoted there by S_i, where i=1,...,T.

4. Time series extraction - using Principal Component Analysis, we exctract the time series s_i from the snapshots as described in our article, Appendix, section 2.

Usage

All the scripts described below can be run on Linux. Assumming, you run the terminal in main project folder, you shall give the execution permission to all the project scripts (e.g. with chmod +x path_to_script command).

To generate the time series described in the Description section, in the main project directory, run

sudo ./main.sh -bp blockchainDirPath -mran minimalRepresantativeAddressesNumber -mid minimalIntervalInDays -mtn minimalTransationsNumber -sdp snapshotPeriodInDays -ewp snapshotEdgeWeightParameter -cn componentsNumber

or, if the number of blocks to be processed has been specified,

sudo ./main.sh -bp blockchainDirPath -bn blocksNumber -mran minimalRepresantativeAddressesNumber -mid minimalIntervalInDays -mtn minimalTransationsNumber -sdp snapshotPeriodInDays -ewp snapshotEdgeWeightParameter -cn componentsNumber

Note that you may be prompted for interaction which should concern providing your home account password or accepting installation or build processes.

Each of the commands above generates the following data:

1. Bitcoin users' graph which is saved in the contractions/usersGraph.dat file. Each line of this file contains the following four parameters: userInputID, userOutputID, bitcoinAmountInSatoshis, timeStamp. The auxiliary ouput file is contracted_addresses.dat. For its specification, see contractedAddresses.sh. If you wish to dump files only, see dumpProcess.sh, and if you wish to create the users graph from the dumped files, see usersGraph.sh.

2. Long-term subgraph saved in the contractions/long_term_subgraph.dat file whose structure is the same as that of the users' graph. In order to generate only the long-term subgraph from the previously created users' graph, see the description in the longTermSubgraph.sh script.

3. Snapshots - saved in separate files (one file for each snapshot, named with the date of the snapshot period) in the snapshots_<number/value> folder. The structure of each line in each snapshot file is as follows: userInputID userOutputID edgeWeight, where edgeWeight is the total number or amount of Bitcoins sent (whether the number or Bitcoin amount depends on the parameter snapshotEdgeWeightParameter from the main command) between the users with userInptID and userOutputID in the period for a given snapshot. To generate the snapshots separately (from the previously created long-term subgraph), see the description in the snapshots.sh script.

4. Time series - saved in the time_series_<number/value> folder. As each time series is defined by the corresponding principal component (see our article, Appendix, section 2 or the description in the pca.py source file), the files containing snapshots' data are named with components' idies. Each such file contains the column of T numbers defining the corresponding time series. To generate the time series separately (from the previously created snapshots), see the description in the timeSeries.sh script.