Trade in Groups

As the year draws to a close, investors typically devote some time to tax planning in order to minimize their bill come April. One way to do this is by taking losses on stocks, and given the terrible year the market has had, there will be plenty of chances to do so. A question in my mind is whether or not this could potentially lead to even further declines in the market, as selling pressure to realize losses before year-end could turn many people into sellers. I turned to some light data analysis to try to figure out whether there is a relationship between the market’s performance in December and the rest of the year.

Using a dataset on the S&P 500 going back to 1950 (I could not determine how long the tax deduction has been allowed), I put the January-November return on the X-axis and compared it to the return for December (Y-axis). The relationship is rather scattered, and has a low correlation coefficient of 0.19:

But the focus I had was only on performance in years when the markets were down going into December, since there would presumably be more tax-related selling. Looking at the relationship only in down years, the correlation coefficient (not a perfect tool, I know) is markedly higher at 0.43, which somewhat confirms my thesis.

To the chart, I added a trend forecast line (pink) as well as a data point (yellow) that shows the expected return based on the inputs. With the S&P 500 down 46% year-to-date, it isn’t pretty: the model calls for an additional 6.3% loss, which would put us back to around 750 points.

Now, I don’t want to represent this model as having excellent predictive values; I’ve never used it before. But I wanted to see if, historically, a model of this type showed some ability to forecast the data presented above. So I sorted the negative return years by when they occurred, and set the forecast model to use only past data in making an estimate of future returns. The estimate compared to the realized return is shown below; two observations – one, the model had a rocky opening when there was little data to work with, and two, it tends to underestimate market volatility (which I did not try to model for). Market volatility is extremely high right now, and worth paying attention to.

What I did focus on is whether or not the model was directionally correct – that is, did the market actually go up when the model expected a positive return, and did it go down when the model expected a negative return? As I mentioned, the model started out on shaky footing when it had limited data, but the last seven predictions have been directionally correct, as have ten of the last eleven. Now, we are in uncharted ground in terms of returns (note the worst January to November return in the dataset was about a 30% loss, and we are down 45% so far this year) and volatility, but overall this reinforces my current dislike for stocks. I continue to look for opportunities in the debt markets and high-yielding alternative assets, but would probably have protective puts under long equity positions.