This segment features a discussion of predictability versus randomness and includes some “relatively” easy math that demonstrates why some trading models are flawed. Dr. Data (Michael Rechenthin, Ph.D.) guides us through the numbers and helps us makes sense of it all. All traders will learn something but those doing backtesting, or thinking about buying a backtested trading system, really need to watch this segment.
Take an individual and have him flip a coin ten times and the odds that they will be all heads (or all tails) is very small, about 0.1%. Now gather 1000 people in a room and have them all do the same and the odds that at least one person will be able to do it skyrockets to 65%. That's the example Dr. Data uses to introduce today's topic.
Applying a trading strategy to historical data to see how well the strategy would have worked is known as a backtest. The danger is that a poorly constructed backtest can produce misleading and overly optimistic results. Dr. Data teaches us to recognize the difference.
The “Skinny on Options Data Science” on September 9, 2015 examined the stock picking “success” of some different firms. Dr. Data displayed two equations. The first showed the low probability of one firm getting 8 out of 10 picks right. The next showed the near certainty of at least one firm doing this when 100 firms are measured.
Dr. Data's dissertation (yes Mike is a real PhD) was in this area. He took market stats from the S&P 500 going back 60 years with the goal of predicting on day in the future. Using Artificial Neural Networks and optimal weightings of over 200 different technical analysis indicators he was able to design an algorithm with over 93% accuracy when measured to the data. That though was the wrong way to do it. He then explained the correct method to use for backtesting and when done that way his results were quite different.
Dr. Data explained the problem. Those interested in more information on this can watch the segment from April 15, 2015, “Prediction with Supervised Learning Models: Problems with Model Overfit”.
Watch this segment of “The Skinny on Options Data Science” with Tom Sosnoff, Tony Battista and Dr. Data to see the important takeaways and to learn when predictability is actually just randomness.
This video and its content are provided solely by tastylive, Inc. (“tastylive”) and are for informational and educational purposes only. tastylive was previously known as tastytrade, Inc. (“tastytrade”). This video and its content were created prior to the legal name change of tastylive. As a result, this video may reference tastytrade, its prior legal name.