Jegadeesh and Titman 1993 Power of their test

Question

I am reading this classic paper(http://www.business.unr.edu/faculty/liuc/files/BADM742/Jegadeesh_Titman_1993.pdf) and got confused by one of their arguments on their overlapping portfolio strategy to test momentum. They claimed on page 68:

To increase the power of our tests, the strategies we examine include portfolios with overlapping holding periods.

I don't quite sure why overlapping holding periods increase the power of their statistical test. Can someone please give an intuitive explanation?

@BAR look at the page number, not the actual number of pages. I am referring to the page that is labeled as 68. — zsljulius, Sep 23, 2015 at 4:19

BAR · Accepted Answer · 2015-09-23 04:36:15Z

1

It is due to parameter variation.

By overlapping portfolios they can better show that their results are not a one-off result that only works given this very specific set of inputs.

Without testing with different parameters (stocks, timeframe, etc), results are liable to blow up given a different input.

That is not to say using parameter variation always guarantees future results, only that it increases the probability.

answered Sep 23, 2015 at 4:20

BAR

1567 bronze badges

Add a comment |

user3264325 · Accepted Answer · 2015-09-22 05:37:38Z

4

Non overlapping periods would make for a far smaller sample

answered Sep 22, 2015 at 5:37

user3264325

5712 silver badges10 bronze badges

1

$\begingroup$ Yes. For example if you are interested in 3 month returns, and you want non-overlapping periods you only get to form 4 portfolios per year. If instead you are OK with overlapping returns and can deal with that in a statistically correct manner, then you could form 12 portfolios per year. But you need to be careful because overlapping returns are tricky to handle. $\endgroup$
– nbbo2
Sep 22, 2015 at 13:29

Add a comment |

zsljulius · Accepted Answer · 2015-09-22 23:21:38Z

0

I have found a great post here explaining the estimation errors with overlapping portfolio construction. http://www.alexchinco.com/standard-error-estimation-with-overlapping-samples/.

I have not finished reading yet, but it doesn't seem to address the problem of power of statistical test.

answered Sep 22, 2015 at 23:21

zsljulius

6503 silver badges14 bronze badges

1

$\begingroup$ There are two issues. In order to do statistics you need independent samples. If you construct a portfolio the weights you assign each period may not be independent because of the overlapping -- but -- the returns may still be largely independent! Which allows the overlapping of data used to calc weights. $\endgroup$
– user3264325
Sep 23, 2015 at 14:10
1

$\begingroup$ The link you link to is another problem - overlapping returns - or lack of independence of returns, which is a big no no. E.g. I include returns from Sept to Nov, Oct to Dec, Nov to Jan etc. $\endgroup$
– user3264325
Sep 23, 2015 at 14:12

Add a comment |

3 Answers 3