fbpx
Tag

EOS Archives - Coinscious

Accurate Crypto Market Data Ultimately Leads to Winning Model

By | Coinscious Lab | No Comments

The world’s most valuable resource is no longer oil but data. [1] This holds true even for the finance industry. The control that financial companies wield over their data gives them enormous power, and the abundance and quality of data they use changes the very nature of the competition. According to Bloomberg, the financial sector is adopting big data analytics to maintain a competitive advantage in the trading environment” [2]. Quantitative- and high-frequency trading are ubiquitous, indispensable tools in current times, and their full value in cryptocurrency trading are being realized. A key aspect that is still often overlooked in quantitative crypto-trading is the quality of the data being used to design sophisticated prediction models.

In this era of cryptocurrency trading, those with the most data of the highest quality will surely win. In algorithmic trading applications, accuracy is one of the best quality indicators of a data source. It determines the execution prices, the model’s behaviour, and the model’s ability to fit the market efficiently and effectively. In the extreme case, high frequency traders care about order-by-order data to simulate precise market-making algorithms. In order to accurately determine what and how much to trade at a low cost, traders desire the finest scales of accurate data with low latency.

Many algorithmic traders incorporate massive amounts of data into their algorithms to create better pricing models and leverage large volumes of historical data to backtest their trading algorithms. Particularly with recent advances in machine learning, the data-driven approach to modelling is being emphasized more than ever before. Market behaviours are learned from black box models that recognize patterns in big data. This means that the accuracy of the data affects what the model learns and predicts. Thus, the more accurate data you have, the better you can simulate execution quality in algorithms.

Available Sources of High Quality Crypto-Trading Data

There are several companies that provide cryptocurrency market data. Kaiko, CoinAPI, and Coinscious are three well-known crypto data vendors. Most of these companies offer live and historical trading, order books, and OHLCV1  (open, high, low, close, volume) cryptocurrencies. However, what remains unknown, until now, is the quality of data these companies claim to provide. Therefore, the key question is: which data vendor has the highest quality data for you to gain a competitive edge?

The Basics

A simple way to assess data quality is to compare the exchange’s OHLCV data with derived OHLCV data. In the analysis below, the hourly level OHLCV data is computed for December 2018 amongst different data vendors. The error rates were measured over eight well-known exchanges: Binance, Bittrex, Bitfinex, Bitstamp, Bitmex, Huobi Global, Okex, and Coinbase Pro.

Figure 1. OHLC error rates for Bitcoin (BTC), Ethereum (ETH), and Ripple (XRP)2. Given that our budget limits us to purchase just one dataset between Kaiko and CoinAPI, we chose the more expensive one: Kaiko’s data

Figure 2. OHLC error rates for OHLC error rates for four alternative coins (ADA, XLM, TRX, ZRX)

Coinscious data proves to be the most accurate among these data vendors for the top 3 coins (BTC, ETH, and XRP). In average, Coinscious data are 38% better than Kaiko’s data, where the relative errors on OHLC are 39%, 41%, 31%, and 37% respectively (see Figure 1). Similar results have also been shown using four alternative coins (ADA, XLM, TRX, ZRX). Surprisingly, even though Kaiko data is accurate for high and low prices, their open and close prices are quite divergent when compared to Coinscious and CoinAPI.

Error In Trading Volume

In Figure 3 and Figure 4, volume error rates over time reveal the dates when the higher error rates occur. The spike in volume error rates occurs in two scenarios; the first scenario occurs when the volume and volume error rates spikes simultaneously, whereas the second scenario occurs when the volume error rates spike, but volume does not. The former can be attributed to increased latency on exchanges as traffic increases, whereas the latter can be attributed to internal server issues.

Figure 3. Absolute error between exchange volumes versus data vendors’ volumes in December 2018 (the lower, the better). The errors were measured for BTC/USD, ETH/USD, and XRP/USD on the top 7 exchanges3.

Coinscious’ error rates remain relatively low compared to other vendors’ error rates. Overall, it is clear that Coinscious data has the lowest error rates with respect to volume data.

Figure 4. Absolute distance error between exchange volumes versus data vendors’ volumes in December 2018 (the lower, the better). The errors were measured for the following alternative coins: ADA/USD at Bittrex, XLM/USD, TRX/USD, and ZRX/USD at Bitfinex.

The volume quality for alternative coins (i.e., altcoins) was also considered. Eight altcoins were randomly selected from different exchanges, including NEO, TRON, XLM, EOS, LTC, ZRX, and ADA. From the figure above, CoinAPI does not perform well on volumes with respect to these altcoins.

Reason For Data Discrepancies Between Vendors

Now you must be wondering, if the exchange provides public API, why would you need to purchase data? Firstly, public APIs have limited histories of information they provide, and unless a trader has stored historical price data, they would need to gather it from a third-party source. Secondly, even though exchanges provide public APIs, aggregating and preprocessing all possible cryptocurrency pairs for different exchanges is cumbersome, and arguably the most tedious step in developing a trading system. This is especially the case as the data receiving intervals gets coarser as the number of requests for data grows. It is for these reasons that the aforementioned data vendors exist.

More importantly, why do discrepancies in the accuracies exist across different data vendors? There are several possible reasons. It could be due to downtimes of exchange APIs. Or, given the thousands of combinations of cryptocurrency exchanges and trade pairs, there exist API rate limits on all cryptocurrency exchanges, and therefore a large number of data collection clients and complicated infrastructure is required.

While many companies are collecting vast amounts of data across different exchanges and coins, the quality of the data may be hidden underneath the quantity of the data. Especially in this era of a data-driven finance world, success and risk can be heavily dependent on the data quality and the data operations environment. Obtaining the right trading tools and hiring talented traders can certainly help, but even then, tools and people cannot guarantee success if the data is flawed. The cryptocurrency finance market definitely could benefit from having more of data quality analysis in order to understand the granular level of datasets and where they can obtain them.

Footnotes

  1. Open, high, low, close, volume (OHLCV) prices.
  2. Given that our budget limits us to purchase just one dataset between Kaiko and CoinAPI, we chose the more expensive one: Kaiko’s data.
  3. Top 7 exchanges include: Binance, HuobiPro, Bitfinex, Bitmex, OKEx, Bitstamp, and Coinbase.

References

[1] “The world’s most valuable resource is no longer oil, but data”. The Economist, 6 May 2017, https://www.economist.com/leaders/2017/05/06/the-worlds-most-valuable-resource-is-no-longer-oil-but-data

[2] “3 ways big data is changing financial trading”. Bloomberg, 5 July 2017, https://www.bloomberg.com/professional/blog/3-ways-big-data-changing-financial-trading/

Litecoin & EOS Surge, With Bitcoin Trading Around $4,000

By | Coinscious Lab, Data Analytics | One Comment

Coinscious Market Report

by Coinscious Lab

February 19, 2019

Biggest
30d % Gain

 Pundi X (NPXS)
+47.06%

Largest 30d %
Gain (Sector)

Payments & Settlements
+6.76%

Biggest
30d % Loss

 NEM (XEM)
25.41%

Biggest 30d %
Loss (Sector)

Data & Information
6.70%

Overview

Released bi-weekly, this report aims to identify broad trends in the cryptocurrency market. In order to reflect the latest developments in this fast-paced and volatile market, the reports plan to focus on metrics derived from a 30-day rolling window of data, this time from January 20, 2019 to February 18, 2019.

Our universe of analysis includes 50 of some of the most widely used and traded cryptocurrencies, and groups them into sectors that reflect similar utility and valuation models. Through analysis of the recent historical performance of individual cryptocurrencies as well as their sectors, we provide a framework for analysis where investors can identify outperforming cryptocurrencies or sectors by comparing their performance relative to peers.

Sector
Constituent Coins/Tokens
Digital Cash BTC, BCH, BSV, LTC, BTG, DOGE, DCR, BCD, DGB
Privacycoins XMR, DASH, ZEC, XVG
DApp Platforms ETH, EOS, ADA, NEO, ETC, XEM, XTZ, QTUM, LSK, AE, ZIL, ICX, BTM, ETP
Resources SC, GNT
Payments and Settlements XRP, XLM, OMG, NPXS, MKR, PPT
Decentralized Exchanges BTS, ZRX, WAVES
Digital Content TRX, ONT, BAT, STEEM
Data and Information IOTA, VET, LINK, REP
Stablecoin USDT, TUSD, DAI

Analysis

Over the past month, 25 out of the 50 major cryptocurrencies that we examined are up from their values 30 days ago. Bitcoin (BTC), the largest cryptocurrency by market cap, recently broke out of the narrow range from $3,600 and $3,800 that it was confined to. It is currently trading around $3,950 at the time of writing. As is often the case, Bitcoin’s move upwards coincided with good news from other cryptocurrencies as well, with 44 out of the 50 cryptocurrencies having positive returns on February 18th.

Outside of cryptocurrencies, the S&P 500 has been performing well, up 3.93% from 30 days ago and closing last Friday at $2,775.60. Figure 1 presents the risk versus return trade-off over the past 30 days by plotting mean daily return versus historical daily volatility for various cryptocurrencies.

Figure 1. Plot of mean daily return against historical daily volatility for individual cryptocurrencies from January 20, 2019 to February 18, 2019. Higher returns at a given level of risk, measured through historical daily volatility, indicates a better investment.

The best performer overall over the past month was Pundi X (NPXS), with a total return of 47.06%. The second and third best cryptocurrency was Litecoin (LTC) and EOS (EOS), with a total return of 42.79% and 23.01% respectively. Pundi X and Litecoin both performed relatively better than the similar cryptocurrencies that make up the rest of their sectors. Pundi X is a token used for payments and settlements, while Litecoin falls into the digital cash category.

NEM (XEM) was the worst performing cryptocurrency, with total losses of 25.41%. Stellar Lumens (XLM) was the second worst performing cryptocurrency, with total losses of 22.56%. Figure 2 shows various performance measures of the nine sectors as well as that of the S&P 500 for comparison and Figure 3 plots the performance over time of each sector. Total returns ranged from -6.70% (data and information) to 6.76% (payments and settlements). There was a dip in all sectors except for stablecoins around the end of January but all sectors have rebounded, making recoveries with varying degrees of success.

Figure 2. Mean daily returns, historical daily volatility, total returns, maximum drawdown, and ex-post Sharpe ratio for each sector from January 18, 2019 to February 20, 2019. Less negative maximum drawdowns and more positive Sharpe ratios are more desirable. The Sharpe ratio is calculated with the 10 year US Treasury bill rate as the annual risk-free rate.

Figure 3a. Price performance over time by sectors that had positive returns between January 20, 2019 to February 18, 2019.

Figure 3b. Price performance over time by sectors that had negative returns between January 20, 2019 to February 18, 2019.

Figure 4 shows the correlation between the daily returns of each sector. As shown in Figure 2, stablecoins continued to fulfill their intended purpose well by maintaining low volatility, mean daily returns near 0%, and a near zero total return of -0.20% over the observation period. Stablecoins had moderate negative correlation with other sectors except for digital content, with which it had near zero correlation. As for the other sectors, they were less positively correlated with each other compared to the numbers we reported two weeks ago. Correlation ranged from 0.54 to 0.92, slightly lower compared to 0.66 to 0.95 previously. The S&P 500 also had near zero correlation with all cryptocurrency sectors.

Figure 4. Correlation between daily returns of each sector from January 20, 2019 to February 18, 2019. Correlation ranges between -1 and 1. Correlation close to 1 indicates a more positive relationship between the pair of cryptocurrency returns and correlation close to -1 indicates a more negative linear relationship. Correlation close to 0 indicates no linear relationship.

APPENDIX A: Methodology

The daily price data of cryptocurrencies in USD at 4:00 PM EST from January 20, 2019 to February 18, 2019 was used for our calculations.

The prices are the volume weighted average price of the cryptocurrency in USD at 4:00 PM EST each day across all exchanges where Coinscious has data. If there was insufficient good quality data on a cryptocurrency’s value in USD, we would instead use the cryptocurrency’s value in USDT and apply a conversion rate to turn it to USD. If data was still insufficient, then we would find the volume weighted average price of the cryptocurrency in both BTC and ETH, then converted both into USD, and finally took the mean of those values. The conversion rates we use at a given time are the volume weighted average price of USDT, BTC, or ETH to USD at that specific time across all exchanges where Coinscious has data.

To analyze performance by sector, the prices of constituent cryptocurrencies was normalized by dividing by the price on January 20, 2019, then averaged. When calculating the daily returns using this averaged normalized price, it is equivalent to if each sector was represented as an equally weighted portfolio of its constituent cryptocurrencies formed starting January 20, 2019 and the returns of the portfolio were calculated. Returns used throughout this report refer to simple returns.

Daily closing price data of the S&P 500 index from Yahoo Finance was also used as a proxy to represent the US equity market. The latest 10 year US Treasury bill rate from YCharts was used for calculations involving a risk-free rate.

In subsequent reports, we may update our universe, sectors, methodology, and analysis to reflect new developments.

APPENDIX B: Terminology

  • Volatility: A measure of the dispersion in the trading price of an instrument over a certain period of time, defined as the standard deviation of an instrument’s returns.
  • Drawdown: A measure of the decline of the trading price of an instrument or investment since the previous peak during a certain period of time. Less negative, less frequent, and shorter drawdowns are more desirable.
  • Maximum drawdown: The maximum peak to trough decline of the trading price of an instrument or investment over a certain period of time. Less negative maximum drawdowns are more desirable.
  • Sharpe ratio: A risk adjusted measure of return that describes the reward per unit of risk. The reward is the average excess returns of an investment against a benchmark or risk-free rate of return, and the risk is the standard deviation of the excess returns. A higher Sharpe ratio is better. Ex-ante Sharpe ratio is calculated with expected returns whereas ex-post Sharpe ratio is calculated with realized historical returns.
  • Correlation: A measure of the linear relationship between two series of random variables, which in the context of finance, can be two series of returns. Correlation ranges between -1 and 1. Correlation close to 1 indicates a more positive relationship between the pair of cryptocurrency returns and correlation close to -1 indicates a more negative linear relationship. Correlation close to 0 indicates no linear relationship.

Disclaimer

The information contained herein is for informational purposes only and is not intended as a research report or investment advice. It should not be construed as Coinscious recommending investment in cryptocurrencies or other products or services, or as a solicitation to buy or sell any security or engage in a particular investment strategy. Investment in the crypto market entails substantial risk. Before acting on any information, you should consider whether it is suitable for your particular circumstances and consult all available material, and, if necessary, seek professional advice.

Coinscious and its partners, directors, shareholders and employees may have a position in entities referred to herein or may make purchases and/or sales from time to time, or they may act, or may have acted in the past, as an advisor to certain companies mentioned herein and may receive, or may have received, a remuneration for their services from those companies.

Neither Coinscious or its partners, directors, shareholders or employees shall be liable for any damage, expense or other loss that you may incur out of reliance on any information contained in this report.