Editor's Note: This article comes fromBlockVC（ID：blockvcfund）, reprinted by Odaily with authorization.
Lead:
Editor's Note: This article comes from
Lead:
High-frequency trading is a programmatic trading method that uses complex computer systems to place orders, enjoys direct data channels with exchanges, and features high turnover and low latency. High-frequency trading once played an important role in the U.S. stock, futures, foreign exchange and other markets, and its trading volume accounted for 50% of the three markets at its peak; Due to the intensified competition with the track and other reasons, the profit level has dropped.
The capacity of high-frequency trading strategies in the digital currency market is relatively small. The size of a single team is generally between tens and hundreds of BTC, and the rate of return and Sharpe ratio are relatively high. Its use scenarios are mainly market making and proprietary trading. Lack of economies of scale for funding. However, through the combination of high-frequency strategy logic and other strategies, effective prevention of market and exchange risks under extreme market conditions will help optimize the combination of quantitative strategies to create excess returns.
secondary title
1. Definition, scale and status quo of high-frequency trading
High-frequency trading refers to the use of high-performance computers to carry out programmed pending order transactions, through large-scale pending order cancellations to capture the slight price difference between the purchase and sale of a certain transaction target, and complete profits in a very short period of time. In the United States, where the finance is highly developed, its Securities and Exchange Commission (ie SEC) gave 5 generalizations about the characteristics of high-frequency trading in an official document in 2010:
1. Place orders using ultra-high-speed complex computer systems
In terms of program implementation, in order to improve the running speed of the algorithm, some instructions will be integrated into the computer system (the instruction interval is usually on the order of milliseconds, or even microseconds), and in the development of high-frequency systems, the running speed is faster The underlying language of C++ is mainly;
In terms of hardware, overclocking CPU, FPGA hardware acceleration, and GPU parallel computing can be used to improve overall performance.
In terms of communication networks, because the propagation speed of electromagnetic waves in quartz is only two-thirds of the propagation speed in air, and the congestion of optical fiber networks will increase delays, high-frequency transactions will even abandon traditional optical fiber communications and use microwave and millimeter wave dedicated communication lines. As shown in the figure below, the green line in the figure represents the microwave lines erected between urban agglomerations with dense fiber distribution (London and Frankfurt, Chicago and New York). Quincy Data has been able to transmit transaction data from the Chicago Mercantile Exchange in Aurora to Cartwright and Scaux in New Jersey with a low latency of about 4 milliseconds.
image description
Figure 1 High-frequency trading is directly connected to the microwave communication network of the exchange Source: Quincy data
2. Use co-location and data channel directly connected to the exchange
The so-called co-location means that the server and other hardware facilities of the high-frequency company need to be placed very close to the host of the exchange. Some exchanges even provide server hosting services for market makers near their own server clusters. In this way, trading orders do not need to be relayed by brokers, and high-frequency trading companies will see the information on the order book earlier than other market participants.
3. The average holding time for each position is extremely short
In addition to low latency, another major feature of high-frequency trading is high turnover, which means that transactions will occur very frequently and positions will be held for a short period of time to reduce risk exposure.
4. Mass sending and canceling of entrusted orders
High-frequency traders will process and judge every piece of information that appears in the order book, and establish corresponding positions for a large number of pending orders. In some high-frequency strategies, traders will issue some orders to detect other orders (such as Iceberg strategy), or guide short-term price trend changes through a large number of order withdrawal operations to promote price discovery.
5. Basically keep the position closed at the close
Generally, high-frequency trading will close all positions (except holding bottom positions) before the end of a trading day. On the one hand, holding positions overnight will increase the risk; overnight interest)
Since then, through many ECN platforms, people do not need to go through market makers to place orders, and at the same time, order information that can reflect the real situation of the market has also entered the field of vision of investors. It's a shocking change. At that time, even the New York Stock Exchange and Nasdaq had to follow the torrent of history and devote themselves to the wave of electronic trading platform business. After entering 2000, the SEC stipulated that the minimum price change unit was changed from 0.0625 US dollars to 0.01 US dollars. The increase in price levels made there be more room for batch order cancellation between sell 1 and buy 1.
With the development of high-frequency trading, many traditional giants in the financial industry, such as Goldman Sachs, JP Morgan, Merrill Lynch, etc., have entered the field, and the industry has a group of competitive legendary companies and hedge funds - led by Simons, the father of quantification Leading Renaissance Technologies (Renaissance Technologies), Virtu Financial, which is known for its speed (only one day of losses in 1,485 trading days from 2009 to 2014), Citadel Securities, Two sigma, Jump Trade Co, GETCO, etc.
Since then, high-frequency trading has ushered in a golden decade of rapid development in the US stock market. As shown in the blue histogram in the figure below, it only accounted for about 10% of the trading volume of the US stock market in 2005 and gradually developed to account for 61% in 2009. In the futures and foreign exchange markets, according to the Chicago Mercantile Exchange (CME) and According to the 2009 data released by the Electronic Broking System, high-frequency trading also contributed at least 50% of the trading volume. The orange histogram in the figure represents the situation in the European market. Similar to the United States, high-frequency trading accounted for only 1% at the beginning, and it accounted for 38% in 2010.
image description
Figure 2 The proportion of high-frequency trading transactions in the US and European markets from 2005 to 2014
However, since 2009, when high-frequency trading accounted for the peak of trading volume in major markets, its trading volume and profitability have shown a downward trend. According to the figure above, the proportion of high-frequency trading in the U.S. market has dropped to 54% and 56% in 2010 and 2011; the European market is slightly delayed, and the scale of high-frequency trading in 2012 dropped to about 35%. According to Tabb Group data, the full-year profit of the high-frequency team in 2016 has fallen from the peak level of 7.2 billion US dollars in 2009 to 1.1 billion US dollars. In addition to the decline in the daily average volatility of the market, the author speculates that the decline is mainly caused by the following reasons:
The profitability of high-frequency trading has attracted a large number of financial technology teams to enter the field, making the track increasingly crowded and profits gradually divided;
Some high-frequency strategies threaten traditional asset management companies and hedge funds that believe in value investing or fundamental analysis. In addition, due to the introduction of domestic policies in the United States to restrict some high-frequency strategies that "obviously damage" the market, they are further suppressed;
The overall decline in market liquidity and trading volume due to deleveraging.
The high-frequency team has explored the European and American trading markets so far, and is gradually turning its attention to the "untapped" Chinese market. However, the trading rules and structures of the two markets are quite different. my country's a-share market implements a T+1 trading system (different from the T+0 of the US stock market), which makes it impossible to complete the operation of opening and closing positions within a day. However, with the first trading open-end index fund (Shanghai Stock Exchange 50ETF) listed on the Shanghai Stock Exchange in 2004 and the opening of stock index futures trading in China Financial Exchange in 2010, the high-frequency team began to trade in the ETF, commodity futures and stock index futures markets. It is not difficult to see from the previous development history that this type of trading strategy is deeply influenced by the trading structure and system of the market where it is located, and its domestic development will be restricted by the following conditions:
The granularity of transaction data is not enough. Domestic exchanges will only provide TICK-level data, and cannot obtain more detailed information on order orders and order books.
The domestic regulatory policy is unclear, and the transaction behavior that is suspected to guide the price trend, such as placing and canceling orders in large quantities, is strictly controlled.
At present, the development of high-frequency trading in the Chinese market still has a long way to go.
secondary title
2. High frequency strategy classification and profit model
The reason why high-frequency trading can achieve profitability is the statistical logic behind it is the law of large numbers (law of large numbers), that is, when we repeat a large number of experiments, the distribution of the results tends to a certain fixed value. In high-frequency trading, when the probability of each profit is greater than 50% (the winning rate of passive market-making strategies can even reach 80%), even if the profit of each transaction is meager, but in a large number of closed-loop transactions (buy and then sell) out, the total position remains unchanged), the expected return is positive. Different types of high-frequency strategies have different profit methods. The following will introduce several mainstream high-frequency strategies and their specific profit models.
1. Passive market-making strategy
The core idea of this strategy is that, according to the principle of reverse selection, the market maker inserts the buy order and the sell order at the front of the entrusted buy order sequence and the entrusted sell order sequence to become a new buy one and sell one. Assuming that the two orders can be traded within a certain period of time, the profit will be realized without changing the position.
The picture below is a schematic diagram of the order book. The real situation is often more complicated than this. The market-making strategy will comprehensively determine the price at which to open a position based on the information of ten (or more) buys and sells in the order book. Only five positions are left here for illustration. As shown in Figure 1, the bid-ask spread is relatively large at 100.2-99.5 = 0.7 yuan when there is no passive market-making strategy running. At this time, if the market is determined to be a centrally volatile market, the high-frequency strategy will open a position and insert orders between sell one and buy one, as shown in the red order in Figure 2. High-frequency trading has a huge advantage over manual traders in terms of speed, so if a manual trader cancels his order at the buy 2, 3/sell 2, 3 position and then jumps to the front of the order sequence, The passive market-making strategy will also cancel the order. Based on the principle of reverse selection, the order will be placed again after the price is lowered, and the market-making order will always be kept in the direction closest to the price formation, that is, at the forefront of the order sequence.
Passive market-making strategy is more suitable for the central shock market without obvious breakthrough trend, but once the market breaks through unilaterally, it will cause the position of the market maker to shift, either the bottom position is sold at a low price and cannot be recovered, or it is a buy order After the transaction, the price continues to fall and the position is covered. At this time, the strategy must stop the loss in time. As shown in the figure below, when the 15 lots of sell orders under 100.1 in Figure 2 were completed, the price still went up and directly knocked out the first to third sell orders. After re-judging the market information, I learned that the 15-lot buy order received at 99.7 yuan was hopeless and decisively stopped the loss. Cancel the red buy four order in Figure 3, and directly eat the 15-lot order of sell one 100.45 yuan to complete the cover, which is regarded as a stop loss trigger. Generally speaking, the winning rate of passive market making strategy can reach 80%, and the probability of stop loss is 20%.
Passive market making is one of the main application scenarios of high-frequency trading at present. When the price fluctuates within a range (when there is no obvious trend), the high-frequency market maker will insert a pair of buyers and sellers respectively between the buy one and sell one orders. More favorable quotations, thereby prompting transactions to occur, that is, injecting liquidity into a certain security by narrowing the spread. Although the method of earning a small price difference in the market is meager, the winning rate is high, and considerable profits can be achieved in the case of high-frequency transactions; in more cases, the profit of the market-making strategy comes from the exchange facilitating this transaction. The liquidity reward provided by the transaction-return a part of the handling fee as a commission.
2. Directional strategy
Unlike the relatively stable market environment in which passive market-making strategies operate, trend strategies are used in price reversals or breakthroughs. When the high-frequency strategy detects the external information of the market and knows that the event-driven market is coming, combined with the handicap data, the position will be established in advance, and when the market responds to the event and it is reflected in the price, it will complete the profit-closing understanding. As shown in the figure below, suppose we judge that a downward trend is coming (it can be seen that the number of sell orders is large and dense), so we sell 100 lots of market orders first to cancel all the buy one and buy two in Figure 4, and the remaining 1 The lot is placed in a Sell 1 position at 99.85. Because the market is slow to react, we pre-establish short positions and wait for the price to fall. At this time, when other traders see that the downward trend has been completed, they will follow and reduce their positions. When the price drops to the position of buying four in Figure 5, the high-frequency strategy will take back the 100 lots sold at the high level at the position of 99.4 yuan, and make a profit Close the position. Of course, if you make a wrong judgment on the market, you also need to stop the loss in time, that is, quickly buy back the short position and use the upper sell order to cover it.
In addition to the directional strategy triggered by events, there are also two strategies: trend triggering and command preemption. The order preemption strategy has another name in the academic circle - predatory algorithmic trading, that is, before the high-frequency strategy opens a position, it will continuously send small orders to test whether there is an iceberg order in the handicap. For continuous large-amount buy/sell orders, technical means will be used to build positions before the large-amount orders are completed, and after the price rises, the chips will be distributed at high prices to large-amount orders that have not been successfully placed before. Trend triggering strategy, also known as spoofing trading, is easily linked to market manipulation and has become the focus of supervision and investigation. The specific method is to place a large number of sell/buy orders, and place a small number of buy/sell orders at low/high points. When other traders in the market see the order book information, they will be affected by fear or greed and rush to sell. The trading behavior of selling/buying, so that the small orders that have been ambushed in advance can be traded, and the position can be closed with a profit after the trend returns to normal.
The core idea of the directional strategy is basically to determine the short-term high probability price fluctuation direction after studying and judging the order flow information or specific events, to open positions in advance with the advantage of speed, and to close the positions after the price fluctuates to the expected point. The reason why this strategy can be profitable is that the research on the microstructure of the trading market (order book information), event grasping, and execution speed far crush other participants.
Similar to order preemption and spoofing transactions in the trend strategy, the structural strategy also makes a fuss about the trading mechanism to seize the opportunity, which is more or less suspected of undermining the fairness of the trading market. As mentioned above, some high-frequency companies will place servers and other hardware facilities very close to the host of the exchange, and some exchanges even provide server hosting services for market makers near their own server clusters. In this way, trading orders not only do not need to be relayed by brokers, but also high-frequency trading companies will see the information on the order book earlier than other market participants. In addition, there used to be structural strategies such as "no review channel" and "lightning order", but they were banned by the US Securities Regulatory Commission around 2010 because they were too disruptive to market fairness.
4. Arbitrage strategy
Arbitrage can be roughly divided into cross-market arbitrage and cross-asset class arbitrage. In the U.S. stock exchange market, the same stock can be traded on different exchanges. When there is a slight deviation in the price of the same asset on different exchanges, it will be quickly captured by the high-frequency team and arbitrage completed. More common are ETF arbitrage and stock index futures arbitrage, which also have very high requirements for speed, and the transmission of instructions is on the order of microseconds or even nanoseconds.
secondary title
3. Advantages and risks of high-frequency trading
1. Advantages of high frequency strategy
For funds that allocate high-frequency strategies to some assets, the profits generated by them are an important part of the excess returns, and will not conflict with other investment strategies, and play a role in hedging risks. From the microscopic level of trading, high-frequency trading pays so much attention to the research of order book information flow that high-frequency strategies can guide the price trend in a short period of time, make correct judgments and complete profits; combined with the speed of its own hardware Advantages can often be preemptive and complete the opening or closing of positions before market participants.
For the overall market, the contribution of high-frequency strategies mainly includes the following aspects:
By reducing the price difference and active market, the desire to trade can be stimulated, which can increase the real trading volume of the market to a certain extent.
Provide short-term impetus for price fluctuations in the trading market, and can cover the gap so that the price will not deviate too far from the normal range in an instant.
Improve the overall transaction efficiency of the market, catalyze the iteration of computer software and hardware technology and feed back the financial industry.
2. Risks brought by high-frequency trading to the market
As early as 2012, an oolong finger incident in the United States caused huge losses to Knight Capital, the former industry leader. During the upgrade of the high-frequency trading system, the technicians did not upgrade a certain server, which caused it to send millions of abnormal orders to the market after the market opened on August 1, involving more than 150 stocks and triggering a circuit breaker mechanism. Trading of some stocks was temporarily suspended. Afterwards, the New York Stock Exchange only canceled the trading of 6 stocks on the day, and the remaining 140 stocks failed to cancel, which eventually led to a loss of 460 million U.S. dollars for Knight Capital, which turned into operating difficulties and was acquired by GETCO.
Since the development of high-frequency trading, it seems that it has been arousing people's reflection and criticism of the systemic risks it may bring. Buffett and Munger, the heads of Berkshire Hathaway, have publicly stated that high-frequency trading is like "rats in the barn", criticizing it for not bringing liquidity to the market, but only bringing transactions quantity. The rest of this chapter will discuss the risks brought by high-frequency trading to the market, the relationship between trading volume and liquidity, and other issues one by one.
image description
The SEC conducted a comprehensive investigation into the flash crash afterwards, and the conclusions drawn can better help us understand the contribution of high-frequency trading in providing liquidity. As shown in the left figure below, the red curve represents the change in the trading volume of the S&P 500 index futures on the day, the blue curve represents the change in the futures price, and the right figure represents the change curve of the number of pending orders. We can clearly see that at 14:45 when the flash crash occurred, the trading volume surged but the market depth plummeted, and the entrusted buy order even dropped to 0. Under the blessing of the market-making strategy, why the market will be vacuumed, What about instant liquidity?
image description
Figure 4 S&P 500 Index futures trading volume and price chart Source: Bloomberg
image description
In May of that year, under the shadow of the European debt crisis, the capital market was tense and sensitive. When an institutional trader decided to place a short order of 75,000 lots for hedging, an unexpected flash crash happened. As usual, due to the existence of high-frequency market makers, orders for such "small" positions will be accepted in full. After judging the order book information and market sentiment, the high-frequency strategy on the spot can decide to adopt the plan of "quotation without transaction intention", or directly adopt "no market making" in order to avoid the market-making position being locked up due to a sharp drop in the stock price strategy. The departure of market makers caused the liquidity of the market to dry up instantly, and the depth became worse. A small number of sell orders could also smash several positions of buy orders, and the stock price plummeted instantly. So far we can see that the target with large trading volume does not necessarily have greater liquidity and real trading demand. The role of high-frequency trading in the market is to provide trading convenience such as smaller spreads for real demand, rather than price root cause of the decline. As for whether high-frequency trading has a boosting effect on large price fluctuations, there has been controversy in the academic circle.
Brogaard selected the order book data of 279 large-cap stocks from Canada's second largest trading platform Alpha Exchange from 2008 to 2012 as the research object. At the same time, in order to describe the market situation more accurately, he selected similar data from Canada's TSX Exchange as a reference. The following two pictures show the situation of high-frequency market makers entering the platform one after another in the past three years. The picture on the left shows the number of merchants that provide market making for each stock during the sampling period, and the picture on the right shows the date distribution of high-frequency traders entering the Alpha platform.
image description
image description
Figure 7 Distribution of dates for high-frequency traders to enter the Alpha platform
image description
Figure 8 Comparison of bid-ask spreads of the same target in two exchanges
So far, the outside world still has mixed opinions on high-frequency trading. While people are surprised by its investment scale and profitability, they are also afraid of the "black swan event" it may bring to the system again. Although every instruction and algorithm of high-frequency trading is completed by machines, how to improve risk control capabilities, reduce system loopholes or think about changes in trading systems and trading structures is what humans should do.
4. High-frequency trading in the encrypted asset market
In order to improve the liquidity level and trading volume, digital currency exchanges usually provide a variety of subsidies and preferential policies to high-frequency trading teams, such as extremely low handling fees or even negative handling fees, and support for high-frequency trading under the exchange server cluster. The market maker provides services such as co-location and lifting API transaction frequency restrictions. For smaller exchanges, it is also necessary to hire a high-frequency market-making team to provide liquidity services for the exchange. However, for teams that are profit-oriented with high-frequency strategies, general trading venues will also choose large trading platforms with better liquidity and more accumulation of retail traders for strategy deployment.
image description
Figure 9: List of handling fees and delivery fees of digital currency exchanges
Figure 10 Digital currency exchange high-frequency market maker fee discount
In addition, the failure of exchanges in the case of violent market fluctuations is also a major hidden danger of high-frequency quantitative strategies. When extreme market conditions appear, high-frequency strategies generally take measures to cancel orders to avoid position shifts caused by excessive price fluctuations and ensure position balance. Due to the cancellation of orders by high-frequency market makers and arbitrage teams, the liquidity of the exchange market will be emptied in an instant, resulting in greater price fluctuations, and higher slippage for other traders, further exacerbating price changes. The emergence of a large number of trading orders and liquidation liquidation will also cause the exchange server to be overloaded or down, resulting in delays in API queries, transactions and other orders. Minor delays may have little impact on low-frequency quantitative strategies, but API delays are more important to high-frequency strategies. The impact is huge, which will cause high-frequency strategies to execute a large number of loss-making orders incorrectly in a short period of time. Even sometimes, API failures will cause the trading strategy to fail to place orders, cancel orders, and execute transactions. Unbalanced positions between multiple exchanges will cause risk exposure, and the allocation of positions will further reduce the strategy's income, which may eventually cause huge losses in the strategy.
epilogue
Figure 11 The market fluctuated violently on March 12, resulting in a large slippage Source: skew
References:
1.Will high-frequency trading practices transform the financial markets in the Asia Pacific Region? Kauffman，Hu and Ma
2.High-Frequency Trading Competition，Jonathan Brogaard， Corey Garriott February 16, 2018
