Transparency and market efficiency : bitcoin trading

Figure 1: mid-quote price BTC/USD on June 23rd 2016

The rise of block-chain and crypto-currency seems unstoppable. I am bombarded by articles, opinions and marketing on how revolutionary this is. I agree! I am excited by how crypto-currencies can reduce frictional costs in monetary transactions. How block-chains decentralize compute and does away with rent seeking intermediaries. And in general the end-client benefits from a truly shared database (or ledger).

But to understand this revolution, I chose to start exploring the world that I somewhat understand — the marketplace on which financial assets get traded. As is the case with other financial markets, the market for crypto-currencies is highly fragmented with multiple venues offering crypto-currency pairs for trading with different business models. Most of the marketplace operate on an exchange like model with full visibility into the order book — akin to lit markets in equities.

Unlike other financial markets, these are completely unregulated markets. Modern equity markets by contrast are highly regulated by securities laws that have evolved (or perhaps accumulated) over decades. These range from investor protection (best execution and order protection rules) to market integrity rules (pre/post trade transparency and order-exposure rules). Even OTC markets such as fixed income and foreign exchange (FX) are subject to some form of these regulations. These highly regulated markets are as fragmented as the crypto-currency markets and have a more complex market structure (participants, technology and regulatory framework). I explore how crypto-currency markets fare in terms of market efficiency when compared to regulated financial markets. Especially given that although unregulated, they are highly transparent in that limit order book data is available freely in real-time. Perhaps these markets can demonstrate that the best form of regulation is transparency. Or as eloquently stated by justice Louis D. Brandeis:

“Publicity is justly commended as a remedy for social and industrial diseases. Sunlight is said to be the best of disinfectants; electric light the most efficient policeman.”

Market efficiency is obviously not just a function of transparency and the measures below are rather simplistic. Nevertheless, I thought it would be an interesting exploration. But I have to be modest in my ambitions for this first post. The goal of this post is simply a superficial exploration of data from a single marketplace on a single day.

First a quick thanks to folks at fstream.io, Bob Tiernay and Anthony Cross, for providing the code for marketplace connectivity and order-book reconstruction. Access to full-order book will power a much deeper analysis to come.


Execution costs

Table 1: The traded volume and top-of-book changes to depth and price (quote) for Bitcoin/USD on a single market (GDAX) for a single 12-hour trading session from 8.00am to 8.00pm on June 21st 2016. This is a prior to brexit and Bitfinex hacking and as far as I know, there were no specific market events.

Table 1 show key statistics for quoted and effective spreads and figure 2 shows the intraday variation in spreads. The mean quoted spread is 24 basis points (bps) or 17 bps when weighted by time. To put this into perspective, retail EUR/USD spread ranges from 1–1.3 bps on retail FX platforms such as OANDA[1]. The time-weighted mean quoted spread for TSX Composite (liquid Canadian equity index) is approximately 12 bps (and effective spread of 11 bps). In addition to quoted spread, some crypto-currency marketplaces impose a maker/taker rebates/fees. The liquidity provider (maker) is given a rebate and the liquidity consumer (taker) pays a fee. The marketplace earns the spread between the make and take fees. In this case, in addition to the 24 bps quoted spread, the marketplace imposes a take fee that ranges from 10bps to 25bps. This means the total execution costs could be as high as ~50 bps.

Figure 2: intraday spreads

Effective spreads show the transaction cost of executed trades. During this period the mean effective spread was 31 bps (or 38 bps when weighted by volume). This is higher than the quoted spread measures above. Generally, this shows that participants are submitting marketable orders that exhaust liquidity posted at the best price and access more expensive liquidity deeper in the book (walk-up the book). It is also indicative of an active order-flow that is not minimizing execution costs by either breaking up marketable orders and timing execution to when spreads narrow. Compare this to a modern equity market. On the TSX60 index, the ratio of effective spread to quoted spread is approximately 0.83[2] in contrast to the ratio of 1.28 observed here.


Liquidity

Table 2: quoted depth measures

On a time-weighted basis, the ask depth is approximately $2,300 and bid depth is $1,600 (see table 2). This shows relative light liquidity, however, quoted depth is well above average trade sizes. The imbalance (ask amount> bid amount) in depth is observed over most of the day (see Figure 3). An interesting observation is that quote count increases starting 17:00:00 hrs (see Figure 4) corresponding with an improvement in depth and indicative of market making concentrated in a different timezone (outside of the 12 hours trading window I analyzed). I would have expected trading activity to show a similar increase, however, a cumulative plot of trading volume shows a fairly consistent activity throughout the day.

Figure 3: intraday quoted depth
Figure 4: left-hand graph showing cumulative trading volume, right-hand side showing intraday quote count

So what have I learnt ?

In today’s financial markets, data is a valuable commodity. Open API’s are rarity and access to real-time data streams for full order-book re-construction is expensive. In case of equity markets, data fees represent a significant portion of marketplace revenue. Data from aggregators is even more expensive. In this respect, the marketplaces for crypto-currencies provide free and open API’s (at least the larger ones). The lack of standards means that one has to maintain separate code bases for connectivity to each marketplace. But despite these challenges, the markets of crypto-currencies offer a level of transparency and accessibility not found in more mature asset classes. So the relative ease with which I was able to access the order-book was a pleasant surprise.

My initial reaction to execution costs that range from 25 bps to 50 bps was — too high! BTC/USD is one of the most liquid crypto-currency to fiat pair, and the marketplace in this analysis is the second largest venue for trading it. But in absolute terms only about $8M USD was traded with a mean quoted depth ranging from $1600-$2300. So given the low levels of liquidity, these execution costs are actually superb. OTC markets such as those for corporate bonds are much larger, yet show similar execution costs [3]. These assets are still mostly traded on notoriously opaque inter-dealer systems. Perhaps it is above noted transparency in crypto-currency markets that results into what looks like a relative efficient market.

In this work, I have spent more time connecting to and collecting order-book metrics than I have analyzing the data. But it has been insightful (to me at least) to look at some of the basic market quality metrics. I plan to collect a larger data-set from multiple marketplaces to see how well these findings generalise. Feel free to contact me if anyone wants to collaborate.


References

[1] For recent spreads on major currencies see https://www.oanda.com/forex-trading/markets/recent

[2] Impact of Dark Rule Amendments (pg. 33): http://www.iiroc.ca/Documents/2015/d215afed-a01e-453d-8f24-bd8ed2b948bf_en.pdf

[3] Market-making and proprietary trading: industry trends, drivers and policy implications (pg. 14): http://www.bis.org/publ/cgfs52.pdf