Q: How did you find out about Quantiacs?
Jenia [Timeroot's mentor] mentioned it to me. We are doing research together – physics problems, totally unrelated – but he mentioned at some point that he was searching some term and Quantiacs came up. He started looking at it and it seemed pretty cool. He was trying that and it seemed like fun. He showed me the rankings, models, and stuff. I said, it seems cool. Ok, I try it out, too. A couple days later I tried it out and yeah, it was fun.
Q: Can you remember your lowest low and you highest high during Q2 Competition?
The best experience was when I was having an idea like some component I could add to the code that would improve it. I think it was something like hedge risks between different sets of things and I was doing some math behind it. Finding a nice clean answer to it made me pretty happy. So that was the highest part. The lowest part probably was: I had something and I was trying to optimize some parameters. For the most parts it was all continuous and fine but occasionally some little part acted discontinuously. I was using SciPy
trying to optimize these parameters and it would just freak out because some of it would be non-differentiable and suddenly I would lose all my results.
Q: What do you think is the biggest hurdle to get started and be successful on our platform?
The biggest hurdle is, you are just handed this block of data and figuring out what do I want to do with it. You look at a graph and have an intuition and like a lot of CS problems. The hard part is taking what you actually would do yourself and putting it into a program. Describing it mathematically, that’s the hard part.
Q: How did you overcome this hurdle?
That’s a good question [laughs]. Hmm, I guess playing around. Maybe I wasn’t super sure what I do with the graph myself anyway. So, just getting different things to program. Well, I know, I put a lot of faith in the computer to figure out the parameters itself. I just tried a general flexible model and then tune the parameters to work.
Q: What did you learn during Q2 and what were your most important takeaways?
I learned that finance actually has a lot of pretty cool math behind it. More than I probably gave it credit beforehand [laughs]. I learned cool things about how the stock market works. I’ve learned decently cool things like high-frequency-trading versus daily trading versus market making. I’d heard all these different things about how there’s these interesting problems related to finance. But I wasn’t aware of the diversity of what happened. That’s probably what I learned since then, how some things in finance work.
Q: What did your friends say when you won the Q2 competition?
Most of them thought that it was pretty cool that I had one million Dollars invested in my code now. A couple of them, I think, thought that it was almost unsettling that I just randomly wrote this code over a couple of weeks and suddenly there is a bunch of money on it. A couple of my best friends thought it was funny because they know how much of an abstract mathy kind of guy I am. The fact that I was doing something like finance, they found was not what they have expected from me. But they all thought it was really cool.
Q: What is the most important character trait for a quant developer and why?
I think, being able to trust the numbers if that makes sense. I guess, a lot of things people might try to do is based on intuition. People will use their emotions and look at graphs -- people who are doing manual trading. People look at graphs and try to get a feel for the market. But there are so many biases and studies how biased we are to different things. Even if we get it wrong we find a way to rationalize it with ‘oh, such and such a thing happened’. Putting those things aside and putting faith into numbers and checking statistically if this works or not, that is what you have to do to get reliable results.
Q: Can you give three simple tips to people who are new to algorithmic trading?
Keep in mind – this is data science – like in any science it is useful to take multiple data points. So, don’t just look at one day’s data. Feel free to use averages. There is a whole bunch of different stuff in the industry: moving averages or wavelet averages will do well.
Testing your theories -- if you think that such-and-such a stock is falling or such-and-such a trend that has a positive momentum and always regresses back to a certain thing. Before you try that on every stock you maybe try it on just one and see if it is working here.
You don’t have to judge the accuracy of your claim just on how your algorithm is doing. You can take the data, take it somewhere else, and check in what ways am I wrong. Is there an easy way to filter out the cases where I’m wrong? Maybe its going to have positive momentum and then only a few times its going to drop suddenly. If you can figure out when that is, then you’ve got a good algorithm. Those are two big ones.
Thanks to Timeroot for taking the time for this interview. We hope to see him competing in Q4 and other competitions.