Thursday, September 26, 2013

Why Seasonal Data is Both Irrelevent and VERY relevent

Statistics shows that the larger the sample size as a percentage of the population, the more confident you can be in whatever conclusion you draw. However, the more specific the information the more useful the information.

Seasonal data follows the laws of statistics in that it requires a large number of samples to allow you to draw a conclusion, but how actionable is that conclusion? For example, if seasonal data of an individual stock says 9 out of 15 times (60%) a particular stock outperformed the S&P during the month of October, perhaps you can only be 52% confident that this particular stock will outperform and that the result weren't just due to randomness. Yes you more likely have an edge than not, but not one with substantial confidence. Considering that if you don't have an edge any bet is too large, and if you only have a slight one, only a very small bet is permitted and the fees and commission will reduce that edge even more.

On the other hand, consider sector data that goes back for hundreds of years that contain an AVERAGE of every stock in the sector. If thousands of stocks over 50 years had on averaged outperformed, this is extremely significant and a very large sample size. So if it tells you that there is an edge, with thousands of "trials" you can be much more confident that going forward tech stocks will outperform in October.

The problem being it won't tell you whether or not a specific company like GOOGLE is likely to outperform. However, you can deduce that a stock, selected at random in the tech sector should produce an average tech performance, and give you outperformance over time. Now if you can look at the same seasonal data and also determine that google has a 54% chance of outperforming the average tech stock, then even if you are wrong and it's average, you still outperform the S&P.

Take a scenario where 75% of the time tech outperforms, Even if you select the bottom 50% of tech, you still have 25% of those bottom 50% that outperform the market, and only the other 25% doesn't. So overall you can increase your odds slightly AFTER determining a sector by choosing an individual stock's seasonal data, but you won't have much of an edge just looking at that data. As such, the seasonal data can be almost irrelevant or extremely relevant depending on how you use it. The same is true with most data.

Ultimately you want a large number of samples in similar situations to deliver a combination of both a large sample size as well as a very similar and high quality size.

I love candlestick patterns because of the ability to test millions of them and get a very large sample size, and I feel they are more objective than price patterns. I still like price patterns though. Ultimately, statistically speaking SECTOR seasonality as a guiding principal to focus more on and allocate more towards the sectors with strong seasonal data will provide you with a statistical edge. Then selecting entry criteria in that particular sector measured by candlestick data will add an additional edge, and throw in support/resistance if you'd like and F.A.S.T. Graphs (intrinsic value fundamentals) or other data which perhaps has been statistically tested, and that is where you have an edge on top of an edge on top of an edge. The probability of all of those edges failing and stock under-performing is slim, and that is how you can use statistics to your advantage.

No comments:

Post a Comment