According to the report of the Securities and Futures Commission of the United States, the use of programmed trading modes, such as batch trading, high-frequency trading, algorithmic trading and cross-market trading, has very serious implementation risks. At present, the American market can issue 1222 orders per second due to the use of high-speed automation algorithms. Suppose that the average trading volume of an algorithm is 322 shares and the average trading price is $20. If repeated orders are placed due to the failure of algorithmic trading, 65.438+0.2 million orders worth 720 million US dollars will be issued within 2 minutes, which will not only bring huge impact to the market, but also cause serious consequences.
situation
"Knight Capital Oolong" Event-$440 million evaporated instantly
On Wednesday, August 20 12, within 1 hour of the opening of the U.S. stock market, traders on the new york Stock Exchange (NYSE) felt that the stock moves abnormally, and the prices of some stocks fluctuated abnormally. Just as the market was looking for reasons, KnightCapital, the market maker ranked first in the retail stock trading business of NYSE and NASDAQ 20 1 1, issued a statement saying that there were trading technical problems in the market-making department of the company, which affected about 150 stocks in NYSE. The failure of this transaction led to a pre-tax loss of $440 million for Knight Capital, which caused heavy losses. In two days, the company's share price plummeted by 75%, and it was once close to bankruptcy.
The cause of the accident was that the NYSE upgraded its trading system that day and launched a new trading system. After receiving the notice, the development department of Knight Capital updated the relevant code of the execution system, but it failed to pass the comprehensive test before going online. When the market opened, there was fatal logical confusion in the market-making system, and the order operation of "buy high and sell low" was carried out strangely. This abnormal trading program can't get the confirmation information of the order from the exchange, and it doesn't know that a valid order has been issued. Instead, I think that the order has failed and there is no deal, and I will continue to add supplementary orders. As a result, for the same reason, the additional order also failed, so I continued to place an order. In this way, the higher the stock price is pushed, the more motivated the program will continue to buy, further pushing up the stock price? The resulting chain reaction makes the actual order quantity increase geometrically. After fully automatic trading, some stock prices rose by 100%~200%, and some transactions lost 15~ 17 cents each. Most transactions lasted 5-26 minutes, and the longest transaction lasted 40 minutes. Fortunately, the abnormal situation was discovered by the monitoring personnel of the trading system shortly after it appeared, but the normal management operation could not stop the trading procedure, and finally the trading could only be stopped by physical methods such as disconnecting the network. The direct result of abnormal program trading is an unusually large position, with a maximum market value of $7 billion. After some emergency treatment, the total position at the end of the day still reached 4.6 billion US dollars, resulting in a loss of 440 million US dollars. After a lapse of one year, the "Oolong Incident" of Knight Capital finally came to an end after agreeing to accept a fine of $6,543,802,000 from the regulatory authorities.
This event has a very valuable reference value for quantitative traders:
(1) We must believe that computers will make mistakes, and all kinds of fault protection measures need to be designed in advance.
In the final analysis, the designer of the trading program will make mistakes, so it is necessary to foresee all kinds of extreme situations in advance and provide preventive solutions, including stopping trading on the interface, closing the trading program, cutting off the network connection and so on. If the abnormal program accumulates a large number of positions, it is necessary to consider its liquidation method.
(2) Manual monitoring must be conducted by full-time traders.
The monitoring interface should display complete transaction information and control elements. Traders can check whether the current transaction is abnormal according to actual combat experience.
When conducting equity transactions such as stocks/ETFs, traders should check the data at least once every 5~ 10 minutes due to the slow data release frequency and transaction speed. When conducting futures trading, the inspection frequency needs to be higher.
(3) Trading procedures must be strictly synchronized with the upgrading of trading equipment of the exchange and maintained regularly.
In view of the current situation in China, IT departments should work closely with exchanges and system developers to maintain the integrity of the system.
(4) The trading procedure must monitor the abnormal fluctuation of price and volume, and terminate the trading in extreme cases.
The trading procedure should check the fluctuation, fluctuation range, rate of change relative to the recent price, the change of market trading volume and the change of the trading volume of the system itself in order to control it accurately. Take Knight Capital as an example. Since there is no limit on the price of US stocks, when buying a single order exceeds 1 time, and the trading volume is particularly large, the trading program should recognize this situation and call the police in time.
(5) The trading program must judge the position, including the position of each trading object and the overall position.
Since US stocks are T+0 transactions, market makers generally only need to maintain a small amount of overnight inventory. Due to the trust in its procedures, Knight Capital may not judge the position of the whole or a single stock in the procedures, so it accumulated a large number of positions when the failure occurred. Therefore, before placing an order, the trading procedure should evaluate whether the transaction has a great impact on the overall position of individual stocks/ETFs or futures, and whether there is a risk of exceeding the limit.
(6) The trading procedure should check the relevance of the trading object. Trading procedures should check the correlation between a large number of trading objects, such as whether there are losses in the trading of multiple objects, the quantity and price of losses, and the fluctuation range of trading volume.
situation
Everbright Securities Oolong Finger Incident
20 13 8 16 6 1 1: 05, the Shanghai Composite Index soared within one minute, reaching a maximum of 5.96%. Many heavyweights suddenly paid huge bills, causing the entire stock index and other stocks to rise rapidly, so that as many as 59 heavyweights, including China Industrial and Commercial Bank of China and China Petrochemical, instantly blocked the daily limit and then continued to fall (see the figure below). Until 2 pm, Everbright Securities announced that there was a problem in the self-operated business of the Strategic Investment Department when using its independent arbitrage system. This incident caused many institutions and investors who followed the trend to "lie down" and suffered huge losses, which was called "the Oolong Finger Incident of Everbright Securities".
Everbright Securities Oolong refers to the huge impact on the stock market caused by the incident.
The reason for this risk accident is that the automated arbitrage trading system purchased by the company has defects. The system includes two parts: order generation system and order execution system. During the verification, it was found that the order execution system failed to effectively check and control the amount of available funds when entrusting the market price for high-frequency trading; However, the defects in the order generation system will lead to unexpected orders in certain circumstances.
On the morning of the same day, the traders in this department judged that there was an arbitrage opportunity between the Shanghai and Shenzhen 180 index ETF and the constituent stocks of this index, so they issued the first group of 17 1 constituent stock orders through the arbitrage strategy instruction generation system, totaling 17 1 entrustment, and the entrusted amount did not exceed 2 million yuan. Subsequently, two more orders were issued. Among them, 24 stocks have not been traded, and traders try to use the "re-order" function in the system to automatically fill in the stocks that have not been traded. The system programmer guides the operation, but I didn't expect this function to be verified by the real quotation. Although the system shows that the purchase of 24 stocks is executed, in fact, the program system regards the purchase of 24 stocks as a basket of 24 groups of 180 index stocks, not just 24 stocks, resulting in 26,082 unexpected market orders being repeatedly generated within 2 seconds after 1 1: 05: 08. Due to the defects in the order execution system, the above-mentioned unexpected huge market price declaration and payment of 23.4 billion yuan were directly sent to the exchange's trading host, and 7.27 billion yuan was sold quickly (fortunately, there were no shares of 23.4 billion yuan sold in the market, otherwise the consequences would be even more unimaginable). After discovering this situation, the relevant personnel stopped the operation of the system urgently, and used all their self-operated funds to short more than 7,000 stock index futures contracts in the futures market to hedge the huge stocks they bought (it can be seen that stock index futures contributed a lot, otherwise the losses would be greater), but it still had a huge impact on the China A-share market that day, bringing losses of 65.438+0.94 billion yuan to the company. In addition, on February 24, 2005, the Shanghai No.2 Intermediate People's Court made a judgment on the civil compensation dispute case in which 57 people sued Everbright Securities Company respectively for the "August 65438+2006 incident", and ordered Everbright Securities to compensate the plaintiff for a loss of RMB 4.25 million.
The lesson of this incident to investors is worth summarizing:
(1) The trading system has major defects.
The arbitrage strategy quantitative trading system used this time consists of two parts: the order generation system developed by the company itself and the order execution system purchased from another software company in Shanghai. There is a running-in problem between the two subsystems. There is no fully simulated trading operation to test the effectiveness of the trading system, and the risk can be imagined by directly using the arbitrage system for real trading.
(2) Risk internal control mechanism exists in name only.
The strategic investment department system of Everbright Securities is completely independent of other systems of the company, even not under the monitoring of the company's risk control system. Therefore, the deep-seated reason of this incident is that the multi-level risk control system has not played a role.
At the level of traders: although there are three risk control systems for trading varieties, the opening limit and the stop loss limit, the latter two have not played a role.
Department level: Although the department's firm trading limit is 200 million yuan and the daily operating limit is 80 million yuan, it has not played a role.
Company level: the company monitoring system did not find a huge amount of 23.4 billion yuan to pay the bill. This huge sum of money either used funds from other departments of the company to supplement the generation and execution of purchase orders, or there was no position control mechanism at all.
In traditional securities trading, the response of the risk control system is the fastest in seconds, but it is far from meeting the requirements of high-frequency arbitrage trading. For example, in this incident, the generation time of each order is 4_6 milliseconds, and the risk control system developed by traditional IT technology will bring huge delay, which will seriously affect the order placing speed, which may also be the real reason for the "failure" of each link of risk control.
(3) Early warning mechanism needs to be improved. This incident reveals that the early warning mechanism of the exchange as a front-line supervisor needs to be further improved in the face of abnormal quotation and abnormal number of declaration forms.