Assessing Systemic Risk in the Bittrex Cryptocurrency Market
Michael Kane
Yale University
Outline
What is the Bittrex cryptocurrency exchange?
How do you access the exchange from R?
How diverse are the returns of trading pairs the exchange?
What's a cryptocurrency
A digital asset, maintained on a blockchain
Decentralized
Supply of assets can be limited
Some people want these assets (there's demand)
Most popular is Bitcoin ($127,787,600,969)
What can you buy with cryptocurrencies?
Counterfeit Watches
Fake IDs
Cocaine with a Viking Stamp
Recently more legitimate things
Greek Villas
https://cointelegraph.com/news/15-amazing-things-you-can-buy-with-bitcoin-today
Yachts
https://cointelegraph.com/news/15-amazing-things-you-can-buy-with-bitcoin-today
Others
https://cointelegraph.com/news/15-amazing-things-you-can-buy-with-bitcoin-today
Miniature submarines
Small islands
Morganite (expensive rocks)
Guides for the Uninitiated
What's the ratio of prices between two cryptocurrencies?
Top 10 Exchanges
https://cryptocoincharts.info/markets/info
Why Bittrex?
Lot's of cryptocurrency pairs (290)
Large exchange
RESTful API
The bittrex R package
Simple, low-level client to the Bittrex exchange
Supports the public and private API
Part of the ROpenSci project
Getting a market summary in a few lines of code
> library(bittrex)
> bt_getmarketsummary("BTC-ETH")
$success
[1] TRUE
$message
[1] ""
$result
market_name high low volume last base_volume
1 BTC-ETH 0.07779196 0.076 8840.273 0.07637027 681.4346
time_stamp bid ask open_buy_orders open_sell_orders
1 2018-06-01 18:18:33 0.07630321 0.07637027 3303 5164
prev_day created
1 0.07679785 2015-08-14 09:02:24
Getting an orderbook
# Visualize the usd/btc orderbook in R
library(bittrex)
library(dplyr)
library(ggplot2)
bt_getorderbook("usdt-btc")$result %>%
filter(rate < 2.5e4, rate > 5e3) %>%
ggplot(aes(x=rate, fill=type)) +
geom_histogram(bins = 40) +
theme_minimal() +
xlab("Price") +
ylab("Total Order Size") +
ggtitle("Price of BTC in USDT")
Systemic risk
Financial products are often correlated (cointegrated) and returns may be influenced by common, underlying factors.
A change in an underlying factor can affect many stocks.
The more cointegrated a set of products are, the more susceptible they are to changes in the underlying factor.
If you own 2 stocks whose returns have correlation 1, do you really have 2 stocks?
What if you have a cryptocurrency exchange with 290 cryptocurrency pairs?
L(\lambda, d) = || X - \hat{X}_d ||_F + \lambda d
Penalizing the Dimension in the SVD
Scale the returns, call it \( X \in \mathcal{R}^{n \times p} \)
Offset increase in approximation accuracy with the cost of adding a dimension
What should \( \hat{X} \) and \( \lambda \) be?
L(\lambda, d) = || X - U_d \Sigma_d V_d^T ||_F + \frac{d}{\sqrt{p}}
Penalizing the Dimension in the SVD
The d-dimensional SVD approximation of X.
The square root of one over the number of columns of X.
Two virtual stock in a market of size 290
The corresponding scree plot
What has the range of dimensions been for 2018 so far?
290 trading pairs.
1 - 4
Despite the variety of cryptos, returns are not well-differentiated
See me afterward if you are interested in
- 30-second data from Bittrex