I took a long look at data for about 1000 stocks over a 6 month period, and found a mathematical formula that I believe works well to make money. It does not rely on being able to buy and sell quickly (a 5 or 10 minute lag would not break the system).

I took that system and applied it to all NYSE stocks listed on Yahoo finance since 1962. It works very well in most years (does poorly in 2004, 2008, 2015), and the good years more than compensate for the bad years.

I am aware Yahoo Finance has a survivorship bias. Although I don't have hard numbers to back this up, I don't believe my system would be drastically affected by this. My system is not picking out companies that are more prone to failure than the average publicly traded company. I manually simulated what would happen if a few companies happened to go completely bankrupt immediately after I bought them, and it doesn't affect the numbers drastically due to diversification.

I've spent countless hours looking through my work for mistakes. My girlfriend (a math major) has also looked through it to try to find errors. I've also talked through the basic principles of the system with a few people I trust, and they can't pick out any flaws with it.

As a trial run, I invested $100 using the system. So far (3 weeks in), it has been behaving as I would expect it to behave.

What can I do to further stress test my system? The returns are significantly larger than mutual funds or indexes, so I'd like to put the majority of my expendable money into this. And at the same time, I want to do my due diligence in making sure that I'm not overlooking anything obvious.

  • If this system is automated, read about Knight Capital then make sure your system can't run a muck.
    – quid
    Commented Feb 22, 2017 at 4:04

1 Answer 1


Here are a few things I've already done, and others reading this for their own use may want to try.

If you built the system while looking at Sample Set A, test it on Sample Set B (or the entire population, if possible)

It is very easy to find a pattern in any set of data. It is difficult to find a pattern that holds true in different data pulled from the same population.

Using similar logic, don't look for a pattern in the data from the entire population. If you do, you won't have anything to test it against. If you don't have anything to test it against, it is difficult to tell the difference between a pattern that has a cause (and will likely continue) and a pattern that comes from random noise (which has no reason to continue).

Look at how the system does in historically good and bad years

If you lose money in bad years, that's okay. Just make sure that the gains in good years are collectively greater than the losses in bad years. If you put $10 in and lose 50%, you then need a 100% gain just to get back up to $10.

Plan for Black Swan events

A Black Swan event (popularized by Nassim Taleb, if memory serves) is something that is unpredictable but will almost certainly happen at some point. For example, a significant natural disaster will almost certainly impact the United States (or any other large country) in the next year or two. However, at the moment we have very little idea what that disaster will be or where it will hit.

By the same token, there will be Black Swan events in the financial market. I do not know what they will be or when they will happen, but I do know that they will happen. When building a system, make sure that it can survive those Black Swan events (stay above the death line, for any fellow Jim Collins fans).

Start from scratch

Recreate your work from scratch. Going through your work again will make you reevaluate your initial assumptions in the context of the final system. If you can recreate it with a different medium (i.e. paper and pen instead of a computer), this will also help you catch mistakes.

  • Obviously you want to keep the system secret for now so any specific risk discussion is going to be tricky, have you made a (mental or physical) checklist of what WOULD have to happen for your system to break down? So instead of considering general black (though natural disasters are grey!) swans you might specifically brainstorm what extreme conditions would disrupt your system?
    – Koen vd H
    Commented Feb 20, 2017 at 21:50
  • @KoenvdH Could you give me an example of what you mean? For the sake of discussion, let say my system was to buy every stock that started with the same letter as the day of the month (M's on Monday, T's on Tuesday, etc.). Although it's easy to say that an overall stock market crash would affect the system, I have a difficult time thinking of a specific event that would affect the system specifically, rather than the market as a whole.
    – Jacob
    Commented Feb 21, 2017 at 1:06
  • @KoenvdH A less inane example would be a system where I bought any stock that had increased in price for 5 consecutive days.
    – Jacob
    Commented Feb 21, 2017 at 1:08
  • 4
    In a general sense a market beating system relies on certain presumed market inefficiencies or predictable investor behavior, in your 5 day rise example you probably assume investors become more interested in a stock that has risen for 5 days, thus pushing it up further. In that case you would consider why investors become more interested after 5 days and if it's possible for this interest to vanish somehow.
    – Koen vd H
    Commented Feb 21, 2017 at 15:41
  • @JacobJones "A less inane example would be a system where I bought any stock that had increased in price for 5 consecutive days" Also known as momentum investing. That works great in a strongly trending market (whether the market is trending up or down, because you can choose to go short or long an investment), but works far less well in an undecided market or during sudden market reversals.
    – user
    Commented Feb 22, 2017 at 10:30

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .