Hacker News Reader

Comment by chatmasta | Hacker News Reader

Can somebody explain to me why, if this really works, you would publish it in a blogpost? Shouldn't you be hunting down investments of $X to turn $1.093X?

View on HN

IMHO, when developing a trading strategy it helps to document and share your strategy with others as you'll come to better understand it from the questions and observations others make. No one strategy can or will be successful forever.

Many algorithms stop performing when market conditions (lasting hours, days, weeks, months) change. Having a deep understanding of your algorithm and what makes it successful for any given period of time can better help you make adjustments when needed.

Lastly, this may be where the algo started but not necessarily what they will run in production. It's much more likely to no longer be discussed at this point. Perhaps similar to ideas are worthless, execution is everything for startups.

edit: typo

If an algorithm stops performing after hours or days, it's likely you haven't discovered anything, but are simply seeing the effects of random noise on your hundreds, thousands, or millions of signal possibilities.

One example that comes to my mind could be an algo closely related to the price of another security or index of what have you. At times, this algo could be highly correlative and at other times less so.

I agree with you about random noise. Ultimately I'm just looking for something to make me feel like I'm taking an "informed position." You never really know what's going to happen.

Have their been any studies on the accuracy of backtesting data in predicting actual returns? Comments like yours always appear on these articles... "You can backtest, but you never know what's going to happen!" Well, obviously. But at what point is testing against 12 years of backdata not enough? Is it a matter of understanding exactly what your algorithm depends on and watching for those conditions to fail?

To me, backtests are constructive but like you said, not predictive and perhaps questionably meaningful. Any good strategy using technical analysis will be able to identify a trend in any market. What differs are entries and exits. These factors can be influenced by current events making them tough to model into an algorithm especially with one with a shorter timeframe.

Like selling pickaxes during a gold rush, Quantopian is probably better off encouraging hobbyists to start speculating via their platform.

Yes. Which seems like a better investment: dedication to a platform that certainly will attract speculators and is underserved in general or dedication to a trading algorithm in an arena overpopulated by very smart people competing in a zero sum game -- against you? Both provide the potential for scalable, massive returns, yet in the quant realm, selling pickaxes is a much safer bet and is equally -- if not more -- lucrative. (In the contextually correct parlance, the sharpe ratio is much higher for selling Quantopian.)

I'd be hard-pressed to call this algo as "working". If you look at the PnL, it makes most of its profit from Sept 15 2012 to Nov 15 2012. The rest of the time, it's either drawn down or chopping back and forth. 2 months of gains over a one year period sounds more like noise to me than anything else.

Reply on news.ycombinator.com