The Gateway to Algorithmic and Automated Trading

No Signal

Published in Automated Trader Magazine Issue 40 Q3 2016

NO SIGNAL is a regular column where we examine various snafus in the trading, particularly the automated trading, world. We look at errors in application logic, mistakes by overzealous co-workers, failures in technology and temporary losses of power to both infrastructure as well as craniums. These all make for good stories that everyone can alternatively either learn from or be amused by. If you have a story that you think makes for a valuable lesson or is simply funny in a facepalm moment kind of way, please get in touch with us at no.signal@trader.news. Naturally, we treat all submission with the highest confidentiality. We are only interested in the lesson value, or in some cases the humour value, and not in identifying involved parties.

What are my options?

Option markets are fairly efficient. And they are particularly efficient at extracting money from the uninitiated and the unaware. This story tells the tale of a trading firm that decided that getting an academic with zero practical experience to trade options was a good business strategy. After all, he had spent many years researching volatility models and was happy to expound his understanding of the dynamics of volatility ad nauseum.

With the bright-eyed optimism of a child walking toward towards a lion, the trading firm and their academic in tow set out on the road of volatility trading.

If you haven't traded options before, let me briefly try to explain why trading them is more complex. To begin with, there are the additional dimensions to consider: a stock would have zero dimensions. There are other stocks, but as far as trading this one is concerned, you can only choose to trade this one or not trade it. Technically, there are some companies that have different types of stocks in their capital structure, such as common or preferred stock. But that's pretty much it.

Now, if we add some simple derivatives, like futures and forwards, we have an additional dimension: We can choose what delivery date to trade: We can trade spot for 'immediate' delivery, which will normally be between zero and three days. Or we can extend our reach in time: We could trade a future in three months' time. Or six months' time. Or 10 years' time. You get the picture.

Actually, in most cases all these expiration dates collapse back down to a single point. This is because in many fungible derivatives, such as stocks, stock indices, foreign exchange and bonds, the liquidity is 99% concentrated into the most active expiration month. So everyone basically trades the September 2016 future and it is almost like trading the stock, or bond or currency, itself.

If we add yet another dimension, then we add exercise prices for options. And at this stage, the liquidity will usually no longer be concentrated in just one contract. Yes, the at-the-money options are generally more popular than the out-of-the-money options. But there is still plenty of interest in the out-of-the-money options. And while the one month options are often the most actively traded, there is still significant interest in trading three month, six month and 12 month options. In some cases, people even trade three to five year options.

Wading into options trading for a trading firm that has not traded options before is like stepping into hyperspace; a bit like the characters from 'Flatland' coming out of their two-dimensional sheet world not into three-dimensional space, but into a four-dimensional one. It is going to be weird. And will require some adjustment. One should tread carefully.

Not knowing any of this, but with full faith in the professor, the trading firm mentioned above set out to conquer the world of volatility trading armed with their existing models.

Of course, the back office had never dealt with options before, so the firm's systems were not ready. They also did not understand if and how options were treated in the tax code. And they definitely were not familiar with the conventions in the market place, for instance, how options were traded and portfolios of options were managed.

Armed with forecasts for the volatility of a bunch of single stocks, the firm began to build positions in a number of at-the-money options. This worked well for about eight weeks, as volatility was declining from an earlier short-term spike. And most of the forecasts were pointing towards lower volatility - hardly surprising, as most volatility models simply predict a reversion of volatility to its long-term mean.

An important aspect in equity options trading is corporate actions. Corporate actions are important if you are trading the stock itself, but the consequences of missing corporate actions are usually much less severe. Consider a simple 10-into-1 reverse stock split. Sure, the price is now 10 times higher, but you also have a position that is 10 times smaller, so no P/L impact, right?

In the case of options, there is also some adjustment. Normally the contract size, which is the number of shares a single option is for (normally 100 shares), gets reduced by a factor of 10. So now it will be 100 shares. Also the strike price gets increased by the same factor. So if you had an option struck at 12 USD, you now have an option struck at 120 USD. But the economics remain unchanged.

The professor owned a short position in June 2014 calls with a strike price of 3.00 USD, with each option being for 100 shares. After the reverse split he had a short position in June 2014 calls with a strike price of 30.00 USD, with each option being for just 10 shares. The calls were actually deep in the money before the reverse split, and as a result there was little-to-no actual delta exposure on the position as it was fully hedged. It looked like there was a 1000% jump in the stock, but hey, you can miss that if you are essentially running everything out of an Excel spreadsheet.

The back office did not pick it up because there were no real changes in the liquidity level. The front office did not pick it up because there were no changes in P/L.

What did get picked up though is the new position in the underlying stock, which was now 10 times larger.

Position Before:

Delta
Underlying +250,000 shares +250,000

June 2014
3.00 calls

-2,500
(100 shares each)

-250,000

*Price of Underlying:

4.00 USD

0

Position After:

Delta
Underlying +25,000 shares +25,000
June 2014
30.00 calls

-2,500
(10 shares each)

-25,000
*Price of Underlying: 40.00 USD 0

At the end of the day, the 'auto-hedger' (yes, they actually had one of those) compared its position in the underlying shares, which now totalled 25,000, with the position it needed to have. According to the system, they were still short 2,500 calls with a contract size of 100 shares each, whereas the contract size was in fact now 10 shares each. So, in order to be, ahem, 'hedged', they had to buy another 225,000 shares. Which they did. So now the real situation looked like this:

Delta
Underlying +250,000 shares +250,000
June 2014
30.00 calls

-2,500
(10 shares each)

-25,000
*Price of Underlying: 40.50 USD +225,000

Still nobody noticed because there was no real economic impact for the moment. Yes, you bought a lot more stock, but the mark-to-market of that position is still close to zero because the price of this stock has not really moved. Your margin at your clearing firm would have gone up drastically, but this firm was not the sort of place where big changes in margin requirements were unusual.

This is a big position in delta terms. 225,000 shares of a 40 USD stock is nine million USD. Given that the average stock moves more than 1% a day, your daily standard deviation is more than 100,000 USD. Your 95% VaR is more like 200,000 USD. But that was not enough. Normally stocks of companies that get a reverse split treatment have already had a depressed share price for some time. Which is probably indicative of the economic fortunes of said company. And sure enough, just two days after the reverse split, the company disclosed some massive financial troubles, causing the stock price to plummet by 90%... right to where it was before the stock split. This turned the overall position from a mark-to-market value of zero to one with a mark-to-market value of -9 million USD...

The firm did survive this, although it took them several days to understand what happened. It then took another week for management to really accept that these things could actually happen.

The professor did not survive it and is back to researching even better volatility estimation methods. It is unknown if these new methods incorporate corporate action event handling or not.

What went wrong:

  • Transitioning from linear instrument trading into option trading is a difficult thing to do and has to be executed with great care. Even then it's a move that often fails. The firms that do option trading well, tend to do it from the beginning.
  • As a result: Very limited practical experience in the idiosyncrasies of trading options
  • No proper back office process for processing corporate actions on both underlying and options.
  • No reconciliation of P/L between front office and back office
  • No reconciliation of margin requirements

What went right?

  • Nothing, except, the firm was adequately capitalized to absorb the loss.

Symbology

Processing market data is an arcane field. There are all sorts of caveats, gotchas, pitfalls and tricks you just have to know about that are not really documented anywhere.

One of the areas inside the world of market data that sticks out as a particularly foul area of dark technology is known as 'symbology'. It deals with ways of identifying particular instruments by means of an identifier or 'symbol'. Not unlike how chemists would identify chemical elements, for instance, 'Au' for gold, 'Fe' for iron and so on.

Naturally, because it is finance, we cannot just have one standard. No, everyone has to have their own way of making up identifiers for various instruments. Eurex trades Bund futures under the symbol 'FGBL'. But on Bloomberg, they are identified as 'RX', on CQG, they are 'DB', on Reuters, they are 'FGBL'.

To make matters worse, these things change over time. Trivia: Eurex, which had itself changed its name from 'DTB' in a merger in 1998, (yet another symbol change twist) used to trade the old Deutsche Mark-denominated Bund futures under the symbol 'BUND'. Just like the artist formerly known as Prince, the nomenclature can vary over time.

This story begins with the firm's algorithm trading happily away in the DAX Future contract in 2010. Depending on which data feed you relied on, this symbol was either FDAXU10, FDXU0, GXU0, DXY.U0 and so on. The algorithm was fed from a large market data platform that carried almost every exchange in the world. The firm subscribed to many of them. This is not as strange as it may sound. It is good practice to try and collect as much data in real-time as possible. You never know what you might need at a later stage.

Suddenly, the market seemed to have spiked down: Whereas the future seemed to have traded at around 5000, it suddenly was trading around 25.00. And then it was back to 5000. Nobody even noticed and no trades were triggered because of it.

Suddenly, it plummeted down to 25.05 again, and this time it actually triggered some signals, causing orders to be sent.

Now, in the old days - up to about 10 or so years ago - there would have actually been price filters running behind the feed that would have caused such an enormous (-99.5%!) deviation from the last price. These were common, because in the period between 1970 and 2000, many exchanges were still conducting pit trading where prices were punched in by so called 'pit reporters'. Occasionally, these pit reporters would screw up a price or miss a digit (so a price might jump from 134.50 to 134.05 and back).

However, this was not like in the Prince song '1999'. This was Eurex, the world's most advanced electronic exchange. Their prices did not just go haywire - computers do not tend to make mistakes (or not very often at least).

Looking at the time and sales history of the DAX Future (for September 2010 delivery) showed this:

5023.00

5023.00

5022.00

5021.50

5022.00

25.00

25.00

5022.50

5022.00

5023.00

5022.00

25.05

5021.00

25.04

25.04

25.05

5021.00

5020.50

5020.50

5021.00

25.00

5021.50

5021.00

Okay, that definitely looks wrong. What was happening? It looks like the price was about 1/200th of the normal price. People tried to figure out what was going on. Was it a normalization issue? It is not uncommon for prices to be transmitted in a de-normalized way to save space and therefore save on bandwidth and/or storage. So a price could be transmitted as an integer (50215), which takes less space on the data feed. Then this value would need to be divided by 10 to get the normal price. Did the divisor get messed up and changed to 200? No, the divisor tables seemed to be okay.

The standard response in finance if something is not working is to blame the other party. So, immediately tickets were raised with the data feed provider that they are sending incorrect prices. An hour later came the response: "No, we are not seeing these prices. Must be on your end."

More investigations. No obvious problems. Only once a debugger was attached to the feed handler did the problem become obvious.

The symbol for DAX futures on the data feed was: DXU0. This includes the standard notation for expiration months. Another exchange, a cash equity one, had actually listed a stock with the symbol 'DXU0'. And this stock was trading around 25.00. So its price got simply merged into the prices for the DAX Future September 2010 in the feed handler.

Futures always have symbols including the month code: EDZ8, ESM9, CLH3 and so on. The developers simply had not considered that there might be other instruments in the universe that might have identical symbols to something like 'CLZ12'.

The resolution was to make the instrument symbols on this data feed a combination of the symbol and the exchange code. This combined key would then be used to identify the instrument. For good measure, they also added the data source, so, the feed itself, for forward compatibility with other feeds.

What went wrong:

  • Symbology coding did not include exchange/data source identification. This was ultimately the source of the problem.
  • No price filtering/alerting mechanism on 'uncharacteristic' moves in the price.

What went right:

  • Feed handler was able to be debugged in real-time.
  • Problem fixed in a few hours.
  • Copyright © Automated Trader Ltd 2017 - Strategies | Compliance | Technology

click here to return to the top of the page