Sep

27

I have been looking at S&P data for the first nine months/last three months of the year. What suggestions do the following percentage changes engender?

Year     First Nine Months     Last Quarter
'80           16                  8
'81          -15                  5
'82          -01                 10
'83           16                -01
'84           01                  1
'85           09                 16
'86           11                  5
'87           30                -23
'88           08                  2
'89           24                  1
'90          -14                  7
'91           17                  7
'92            0                  4
'93            5                  2
'94            0                 -1
'95           24                  5
'96           11                  8
'97           28                  2
'98            6                 21
'99            5                 14
'00           -3                 -8
'01          -22                 10
'02          -31                  8
'03           10                 11
'04            0                  9
'05            4                  1
'06            6

Gary Rogan adds:

I stuck the numbers in a X-Y scatter chart, and several features immediately stand out:

1. Any negative return in the first 9 months below -3% predicts a 5-10% return in the last quarter (doesn't happen to apply this year). 10-ish returns in the first 9 months also predict significant positive returns in the last quarter.

2. Low (sub-9%) positive returns in the first 9 months predict a positive return in the last quarter with 100% "certainty" (does apply this year). These returns come in two varieties: very low single digits (the most likely scenario) or large, above 14% returns (somewhat less likely). The large returns are clustered together in 98-99, which makes this seem not to be a likely outcome. Thus the prediction for the rest of this year is low single digits.

3. High (15+) positive returns in the first 9 months have little predictive power.

Mr. Red comments:

After reading your suggestion to do a piece wise regression, I broke down the data into 3 categories, but in a slightly different manner than Gary's. I chose to look at negative returns, positive single digit returns, and positive double digit returns. Interestingly, negative returns and positive double digit returns yielded negative correlations, while the single digit positive returns yielded a positive correlation (even if you decided to take out the years 1998 & 1999). This supports Gary's findings of significant positive returns. However, the one place we would differ is in his analysis of high positive returns. Using double digits, I come up with the highest R^2 at .3468 (while 15+ is far less at .088%).

Question: are we using a big enough sample set? Is 26 years enough data to draw a meaningful conclusion or for that matter subdividing data into subsets of 10, 10, and 6?

I ask as I have been looking at monthly changes in US 10 year yields for the month of October. When looking back from 1981, this data shows very strong negative changes (lower yields, higher prices) on a median and average basis, the same for using 20 years of data. However, I have seen a report that uses data from 1990, or 16 years of data, which shows wildly positive results (higher yields, lower prices). I would argue that the data from 1981 is better b/c of the larger sample set, but could this also be an example of ever-changing cycles, and if so how do recognize the difference.

Putting pencil to paper, I also did a quick regression of the returns you mentioned, and the returns of the last quarter have a negative correlation with the first 3 quarters of the year. Using the regression formula, it would imply a return of 4.565% for the final quarter of the year. However, given the small r^2, it leaves much room for variation.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search