[Machine Learning for Trading] {ud501} Lesson 17: 02-07 Dealing with Data | Lesson 18: 02-08 Efficient Markets Hypothesis

How data is aggregated

tick interval can be day, mintue, ... etc.

 

 

 Price anomaly

 

 

 

 Stock splits

 

 

 

 

 

 Split adjustment

 

 

 

 

 

 

Dividends 

  

 

 

 

Adjusting for dividends 

 

the close price of 2010 watched in 2012 is lower than the one watched in 2015 => due to the adjustment

 

 

 

 Survivor bias

 

one common mistake is that they look at the membership of that universe as of today

then they go back in time and they use that list of stocks for their strategy 

 

 built-in bias => some stocks dies in SP500 and the current stocks in SP500 actually survives during these years!

 

the strategy trained without considering the survival bias => blue curve => used in real life => brown curve (lower performance)

 

=> eager for the survival-bias-free data => not free

 

 

---

---

---

---

---

---

 

 

 Efficient Markets Hypothesis

 

 

 EMH assumptions

 

 

 

 

Origin of information 

 

 

 

 

3 forms of the EMH 

 

 

 

 

 

The EMH prohibits

weak => only technical analysis can't make profit

strong => any one of three types can't make profit

 

 

 

 

Is the EMH correct?  => at least from data, the strong EMH is incorrect