/Working-with-Binance-Coin-Data

Working-with-Binance-Coin-Data

Statistical Methods for Data Science
Fall 2018
Project: Working with Binance Coin Data

Authors

Data Description

  • Our data files contain two primary groups: token network edge files, and token price files. The Ethereum project is a blockchain platform, and our data comes from there. Although Ethereum started in 2015, most tokens have been created since 2016. As such, tokens have different starting dates, and their data starts from that initial date.
  • Token edge files have this row structure: fromNodeID\ttoNodeID\tunixTime\ttokenAmount\r\n
  • This row implies that fromNodeID sold tokenAmount of the token to toNodeID at time unixTime. fromNodeID and toNodeID are people who invest in the token in real life; each investor can also use multiple addresses. Two addresses can sell/buy tokens multiple times with multiple amounts. For this reason, the network is considered a weighted, directed multi(edge) graph. Each token has a maximum token count maxt; you can think of maxt as the total circulating token amount.
  • Binance coin is among the many which run on the Ethereum blockchain and follows the ERC20 standard. These tokens have a limited supply (i.e., token count, which can be found on coinmarketcap.com as circulating amount). Each token may have sub- units. This is similar to dollar in the US. There are around 18 trillion dollars in the economy, and each dollar is divided into 100 cents (subunits). Similarly, there is a token supply, and then there is a subunit for each token. This idea comes from Bitcoin where subunits are called Satoshis, 1 Bitcoin =10^8 satoshis. Coin market cap gives the total supply, but not sub-units, which differ from token to token. Some tokens have 1018 sub-units. That means there can be numbers as big as totalAmount∗1018.
  • Etherscan.io gives these sub-units as decimals, please see the Vechain here: It has 18 decimals, which means each Vechain token has 1018 subunits. https://etherscan.io/token/0xd850942ef8811f2a866692a623011bde52a462c1
  • Price files have no extensions, but they are text based. File can be opened with a text editor and have the following row structure: Date\tOpen\tHigh\tLow\tClose\tVolume\tMarketCap\r
  • The price data is taken from https://coinmarketcap.com/. Open and close are the prices of the specific token at the given date. Volume and MarketCap give total bought/sold tokens and market valuation at the date.

Question 1

  • Find the distribution of how many times a user 1 - buys, 2 - sells a token. Which discrete distribution type fits these distributions best? Estimate distribution parameters.

Question 2

  • How can we create layers of transactions with increasing amounts? This descriptive statistic is similar to bin selection in histograms. For example, we could choose layer1 as those transactions that involve 0.01×maxt in amount. Find a good value for the number of layers and justify your choice.
  • Once you create layers, you can compute a feature in each layer. An example feature is the number of transactions, another one is the number of unique buyers.
  • As each edge has a unix timestamp, it is easy to compute the edge time to a date. For example, 1294226315 is equivalent to 01/05/2011 @ 11:18am (UTC). See the website https://www.unixtimestamp.com/index.php for unix time conversion. R has functions to compute dates from unix time stamps as well. This way, for a given day you can find all layer transactions in that day. For example, you can say on 10/12/2018 there were 25 transactions in layer 1. The price of token on that date was 3.2$. For each day in a token’s history, you can then correlate price vs feature in time.
  • Find an algorithm to compute the correlation of price data with each of the layers (hint: start by looking at Pearson correlation).

Question 3

  • We denote the token price in dollar as Pt for the t-th day. Simple price return is given as (Pt−Pt−1)/Pt−1.
  • Extract at least three features from the token network at day t−1 and create a multiple linear regression model to explain price return on day t.
  • You are free to choose any feature (regressor). Your features can be in terms of numbers (x1 = number of transactions), percentages (x1 = percentage of investors who bought more than 10 tokens), etc. Similarly, you could transform your regressors (x1 = square root of number of transactions).
  • Finding which features to extract from a dataset is called feature engineering.
  • Present your regression model, explain residuals and discuss your findings. Explain the adequacy of your regression model.