Charles Kirk Q & A with Larry Connors: Quantified Trading and Risk, Part 2
Welcome to Part 2 of Charles Kirk’s interview with Larry Connors (to read Part 1, click here). Charles Kirk is the founder of The Kirk Report, one of the most popular, unbiased blogs/websites on trading available to the average retail trader and investor. This interview is one of the most insightful and far-reaching that Larry has ever provided and we hope that you enjoy the conversation.
If you have any questions or thoughts about the interview, feel free to e-mail Larry with them.
Charles Kirk: You’ve tested a lot of strategies and indicators in the past. Thinking back, where did the majority of big breakthroughs come from? Was it something you accidentally discovered in statistical calculating factors, or did you start with an idea, a theory, and develop it from that point on?
Larry Connors: It starts with an observation. Somehow something has unfolded and I start seeing it unfold in a chart pattern combined with some of the indicators that are there. What I’ll then do is make the observation, and before sending it to the research team I’ll hand do the test results. I’ll go back into different periods of time, and I’ll usually know ahead of time where things will likely not work.
For example, I’ll see certain behavior that may be bull market behavior, so I’ll put that behavior into periods where, historically speaking, this behavior runs the greatest risk of breaking down. And if I see it holding up during those periods of time, then I’ll take the next step which is to send it up to the research team. My philosophy is that if I’m going to take up their time, I want to make sure that I’ve observed the behavior already and that it has held up. This way we don’t waste a lot of time going down blind paths.
Kirk: Are there any tools like software you find indispensable in back testing trading strategies?
Connors: This is a common question that we get asked. What Cesar and David have found – and we have no business relationship with this company – is that AmiBroker tends to be one of the better testing platforms. We also know that TradeStation is very good. I know professional traders that use eSignal’s platform. I’m hearing from traders that use Wealth-Lab. I’m hearing from traders who use Ameritrade or Telechart.
For us, AmiBroker tends to do it.
Kirk: Do you still have the notebooks and tests that you did by hand from the early days? Thinking back, is there anything you know now that would’ve saved you a lot of time and energy in the early days?
Connors: Having the opportunity to go through that process was very important. I think that if I had just gone in and started coding things and then started looking to optimize and running all sorts of lines of code, it wouldn’t have been successful. I’ve seen people go down that path, and it tends to become this great, big optimization process with lots of lines of code. I think I can go in and see stock patterns more easily today because I did go in and do things by hand, so I wouldn’t have changed anything that I did back then versus where I am now.
It would have been wonderful to have had Cesar and David there back then. But I think having the process of just having basically a pencil and a notebook and writing down trade by trade by trade gave me a great framework.
For example, one of the more popular programs that we have here is a course called our Swing Trading College. It’s a 14-week program during which we teach people how to swing trade. We meet over a 14-week period of time, week by week, covering all the basis and looking at short-term trading.
What’s interesting is I require the people in the class to do homework each week. And I require them to hand-do it. I ask them not to code it. And what you’ll see is that everyone basically thanks me that they did it by hand. Because they did it by hand they were able to see things that they normally wouldn’t have seen before. If they had simply coded it up and ran the tests they wouldn’t have been able to see that behavior.
So I suggest that for anybody who wants to learn how to trade – do it by hand. Early on, do it by hand. It’s the single best thing you can do.
Kirk: One of the things you said in your book Short Term Trading Strategies That Work, is that many of the edges that you saw in the 1990s are no longer there and markets have become more efficient. So in your view, how does an average person who desires to trade successfully create a trading edge and strategy that consistently works?
Connors: I mentioned a couple of things in the book along those lines.
One was the ability to trade news. The best example that I gave in the book was that of Bloomberg back in the 1990s. Back then, the majority of Wall Street was using Dow Jones as their primary news feed. Bloomberg was scooping Dow Jones many times by two or three minutes. Bloomberg would announce the news and then Dow Jones would release the news as much as two to three minutes later. So basically anybody who had a Bloomberg terminal was getting news before anybody who had the Dow Jones news feed.
For example, if a company announced earnings – now most companies announce after the close, but years ago many announced during the trading day – Bloomberg would have the story first. You could see if a company exceeded the earnings estimate. You could go in and buy the stock and ultimately by the time Dow Jones released it a couple of minutes later it would drive the price of the stock higher and you could sell into that buying.
That doesn’t exist any more. Especially because of the Internet, the assimilation of information has now become almost instantaneous. So that type of edge doesn’t exist any more.
What hasn’t changed – and that was the main point of Short Term Trading Strategies That Work – is that buying into the selling and selling into the buying still works. That hasn’t changed. This is how especially market makers made their money for decades.
Kirk: In overall trading strategy development can you provide some specific suggestions on things to do and to not do that you in your research have found particularly helpful over the years?
Connors: Two things come to mind. If there are too many lines of code, we get concerned.
When you take a look at the things that we’ve published, you are seeing strategies that have as little as three up to maybe seven or eight rules. It doesn’t get much longer than that. If there are hundreds of lines of code, then we don’t touch it. Even if it gets into dozens of lines of code, we won’t touch it. Keeping things as simple as possible is critically important – otherwise over-optimization is potentially taking place.
Also we want to see robustness in the strategy. For example, if we’re looking at a 2-period RSI, looking to buy an ETF with a 2-period RSI below 5 that’s trading above its 200-day moving average – well, it better start showing the edges when the RSI is below 10, and the edges should start to get a little bit better at 8, and they should be even better at 6 and 5.
What we don’t want to see is a strategy that works at a RSI of 7, but doesn’t work with a RSI at 6, then works great at 5 and so on. That would certainly never be something we put our own money in and certainly something that we would never publish.
We also want to see strategies that have held up in bull markets, bear markets, low volatility markets, high volatility markets … We look at a period from 1995 to 1999, where you had a higher volatility uptrending market, and a period from 2000 to 2002 where you had a higher volatility downtrending market. From 2003 to 2007 you had a low volatility uptrending market. Then in 2008, you had a high volatility downtrending market.
You want to see your strategy hold up in all those different types of market conditions.
Kirk: One of the arguments against quantitative strategies is that they’re based upon past market performance. Many people for instance believe back testing is inherently flawed because it cannot properly account for the ever-evolving, changing marketplace. Some things some argue that has been proven yet again with the poor performance of quant strategies in recent years.
Connors: I don’t agree. I think saying quant strategies have had poor performance over the past years is making a generalization. You take a look at probably the greatest quant fund that’s ever been created, the Medallion Fund, run by Renaissance Technologies and James Simons. Their performance remains stellar within the Medallion Fund, and they are mostly mathematicians and people with statistics and physics backgrounds. James Simons’ background is in mathematics. And look at organizations like D.E. Shaw that have consistently put in solid performance over the years. Same background of most of the senior people there.
If you take a look on the CTA side, Toby Crabel has put in an incredible low volatility performance year after year after year using a quantified approach.
Markets do change. What doesn’t change is human behavior. Essentially what we are doing is quantifying human behavior. We are buying at extreme times – we tend to be entering positions when there’s extreme fear out there. When we’re selling stocks or ETFs short we are shorting into positions where there tends to be a great deal of greed in the marketplace.
That is inherent in the marketplace.
Kirk: In your experience where do most quant strategies go wrong?
Connors: There’s probably a hundred answers to that. In a few instances I think they may have over-optimized something. Again, if they are looking at hundreds of lines of code, it’s probably been over-optimized.
Ultimately it depends on the strategy. We’ve seen strategies that on a look-back basis will go through multiple years of success and then all of a sudden start changing. Traders will ask, “What happened?” And it can potentially be that the strategy didn’t have the ability to absorb, for example, high volatility during a downtrending market environment.
Quant strategies can also go wrong if they were depending upon a dissemination of information to gain that edge or depending on a certain type of volatility parameters, that type of edge could disappear.
But overall I would say the number one reason is just optimization. They started wrong, and if they started wrong, they’ll end wrong.
Kirk: Another argument against quant strategies, especially those which are shared with others as you have in your excellent books, is the knowledge and use of these strategies tend to render them less useful and effective. In essence the trading edges disappear with popularity. Have you found this as well? Is there any reluctance in your party or team to share discovered strategy to avoid this risk?
Connors: We’ve been publishing research for well over a decade, and there are a lot of things that are out there that we’ve published that continue to hold up, not only in the testing but also in the trading. What we’ve found often is that the strategies don’t break. It is the people who are trading the strategies that tend to break first. By that I mean that they change the rules along the way, or they don’t take trades at the most extreme points where the pain is the greatest – but where the greatest edges have historically been.
What we’ve seen is that some people will look to change the strategies for one reason or another. They’ll put their own interpretations on the strategies, even though the strategies will tend to have some simple rules to them, they’ll end up changing them a little bit or potentially not taking signals at optimal points. And what that does is that’s not the strategy performing; that’s ultimately the person executing the strategy that is performing – or not performing as the case may be.
That’s an important part of trading systematically: even though strategies are model-driven, ultimately someone has to press that button in order to be able to execute the model. I see that as being the single most important thing in trading: the ability to execute the trades that are there.
Kirk: One challenge for any experienced trader who utilizes quantified strategies as an approach is coping and managing when there are periods where the system does not work as expected or it has in the past. I for one have developed strategies that work really wonderfully for a period of time only to see a strategy hit the wall simply to experience large scale draw-downs. Based upon your experience, how does recognizing the difference between temporary performance lull versus the deeply flawed strategy on that abandoned entirely?
Connors: Another excellent question, Charles. I recommend everybody read – I believe it is available online – the first issue of the New York Times Sunday magazine in 2009. The cover story is called “Risk”.
What they did is go into the guts of the marketplace to see what was happening in 2007 and 2008. They took a look at what was going on in the brokerage firms and in the hedge funds. And what stood out in my mind was how Goldman Sachs reacted to their models.
Goldman Sachs uses a number of different quantitative models. Emanuel Derman – who is a professor at Columbia University now – built some of the original bond models back in the 1980s. I’m relying on memory, but basically what happened is that their models started showing multiple days of draw-downs and multiple days of losses – something they had never seen before. And what their risk committee did was make the decision to pull the plug on the models because they were showing behavior that they had never seen before.
Now understand that Lehman had models. Bear Stearns had models. A number of other brokerage firms probably had the same type of models. After all, many of these firms draw from the same talent pool: MIT, for example, with Professor Andrew Lo who is certainly one of the pioneers in taking technical analysis and quantifying market behavior.
So a lot of the models were the same. What Goldman did that was different than everyone else is that Goldman pulled the plug. They saw that the behavior was different than anything they had seen in the past, and that was the difference here.
Now we’re looking at this in the summer of 2009. And when we look back, Lehman is no longer around. Bear Stearns is basically no longer around. But Goldman is not only around but there was a story yesterday that the bonuses for Goldman partners – having paid back the TARP money – would potentially be the biggest in history.
A lot has to do with the lack of competition – with no Lehman and no Bear Stearns and a number of other companies. But the key take-away is that Goldman was smart enough to stop trading the models.
Again, that article is called “Risk”, and I recommend that everybody read it. If somebody can’t get a copy of that article they can email me at email@example.com, and I’ll forward it to them. It’s a wonderful article.
Click here to read Part 3 of the interview with Larry Connors.
Larry Connors is CEO and Founder of TradingMarkets.com and Connors Research. His two most recent books are Short-Term Strategies That Work and High Probability ETF Trading. Larry also has a daily ETF subscription service. For a free one-week trial, click here.