In a previous post we saw how avoiding being in the market during Earnings publications could be a zero-sum game in the long run. In this post our purpose is to study if it is possible to take advantage of the effect in the stock prices based on the behavior of the prices during financial statements publication dates.
Fundamental researchers try to exploit the information stemmed from balance sheets, income statements or cash flow statements. You can see how the market reacts sharply through extreme returns whenever these statements are published, depending not only on the goodness of the results presented but also on the expectations of the analyzers regarding the figures shown in the statements. We wonder if just the stock returns exhibited during the statements dates perform as a kind of summary or score of the information shown in the statements. Based on these “scores” we may analyze if there exists some inertia in the behaviour of the stock returns in the near future.
We are going to analyze the returns of the S&P 500 components in the presentation dates and we will connect those returns with the future return the following days (for this example we have used a calculation window of 5 future days).
To avoid the market impact in the assessment of the returns, we measure the returns in terms of percentage ranking, both against the rest of the sector the stock belongs to and against the total universe.
To measure the effect in the price of the Financial Statements we calculate the price return from the previous close of the presentation date to the close of the subsequent next date to the presentation date; that way we take into account that the publication may be before, during or after the market open period, so the impact of the information contained in the statements can impact in the publication date or the next date. That said, we are aware that many times the stocks prices show contrarian movements before the financial information is published, moderating the return, however, we consider these market “hesitancy” as part of the insight to value the stock and the effect caused.
In the 2d histograms with 20 bins, we can see the frequency with regards to the publication return ranking and the 5 posterior days to the publication return ranking, separated by sector and against the whole universe (last picture):
The first thing that stands out is the high frequency of the most extreme publication rankings; this was already seen in the past post.
Secondly, most sectors show more frequency in the highest ranking than in the lowest ranking, both in the publication return ranking and in the post return ranking.
Thirdly, as the time frame of the posterior return is longer, the middle post rankings become more frequent than the most extreme ones, as can be seen in next graphs where posterior 20 days return rankings are plotted:
If we focus on the most frequent rankings in publication dates (i.e. the most extreme) we realize that in the case of the highest publication ranking the frequency in the highest posterior rankings (5 days returns) is higher than the frequency in the lowest in almost all sectors.
On the other hand, in the case of the lowest publication ranking, the frequency in the highest posterior rankings is lower than the frequency in the lowest.
If we average frequencies by sector, we see how rankings from 6 to 10 (5 days posterior return) are more frequent when the publication has been positive in its sector (ranking 10), but rankings 1 to 5 are more often when the publication has been negative (ranking 1).
According to these results and the biases seen, we may consider to use the return in the statements publication dates as a signal to complement other signals.
Proof of concept
Let’s test the results seen in the last paragraphs by carrying out an (unworkable) algorithm that selects the stocks from S&P 500 according to their ranking during the statements publication dates. The return each day of the “Positive Sentiment” portfolio is the mean of the returns of the stocks that have shown a return ranking higher than 90% during the next 5 days after the publication (only those days the stocks are considered to be selected in the portfolio). In the periods with no publication statements, the returns are considered 0.
In order to have a benchmark, we consider the same timing as the portfolio in the allocation but considering that instead of allocating to a specific stock, we hold mean of the sector each stock belongs to. For example, if the portfolio holds Apple for 5 days after the publication statement, the benchmark would hold the mean of IT sector during those days.
As you can see in the graph and statistics table, the theoretical portfolio shows higher performance in the long run, although the risk in terms of volatility is higher.
We have seen that there exists some inertia in the behaviour of the stocks after presenting results. Based on the returns during presentation dates we can give a sentiment of the near future returns that could be used as a signal to complement other strategies.
Hope you liked the post and do not hesitate to sharing your ideas about this matter.