Article: Woodseer pattern matching date enhancement

September 25, 2018

Our new approach to dividend date predictions yields more than 80% exact-dates rules on the S&P 500

Woodseer forecasts dividend payments for over 20,000 publicly listed companies using a combination of our proprietary algorithm and a small team of analysts.

As part of the constant system improvements, we’re pleased to announce a major functional upgrade in the forecasting engine.

Because we predict dividend amounts and dates more than 2 years ahead, this enhancement represents a big improvement in our date forecast accuracy.

This is especially useful to

however anyone with an interest in the forward yield of stocks will benefit.

Through examination we determined the optimum forecast method that our trained (human) analysts were following, and computerised this method.

The algorithm we have developed is now overseen by these analysts who are monitoring the output and in some cases, enhancing with their own insight.

This analyst+algorithm approach has allowed us to achieve greater scale through having the tedious and error-prone process of pattern identification and date calculations performed automatically.

Now for the interesting part...

Based on the analysis of many years of dividend history; patterns emerge which are not always readily obvious to the human eye.

Patterns are disrupted by public holidays, which vary year by year and country by country so are especially tricky to factor.

A recent example of a company with a long list of Friday record dates with a few exceptions appeared to the analyst to be predictive of future Friday dates.

However the algorithm had identified a better pattern which matched on a particular date (April 15th) which for the previous 2 years had fallen on a Sunday and a Saturday, so the company had brought it forward to the Friday.

So where a person naturally sees many recent Fridays and would forecast Fridays in future, in fact the better fitting rule is the first business day before April 15th, which applied to 100% of the data history.

The algorithm we have developed traverses 8+ years of dividend payment history, divides them according to their frequency (Monthly, Quarterly, Biannually, Triannually, or Annually) and groups them in their position (e.g Q1, Q2, Q3, Q4).

Then for each of the 4 date types (declaration date, ex-date, record date and pay date) a series of pattern tests are applied to every dividend in the company’s history.

These checks include closest business day before, after or to a particular date, and week number based on calendar weeks plus Excel’s own bizarre week numbering system (which is surprisingly popular) as well as comparisons to other anchor dates (e.g. the ex-date is 2 weeks after the declaration date).

Then in concert with known date rules within some countries, and exchange-rules governing ex-date and record dates, we apply a confidence score.

This confidence is then based on the number of matches to the pattern, the length of history and the recency of any exceptions.

This allows for the pattern to match even with the occasional exception, which is possible whilst still following an underlying rule.

We apply this pattern matching across all major markets globally to produce forecasts on over 20,000 securities at least 2 years ahead.

The analysis reveals that more than 80% of the companies on the S&P 500 are following an identifiable pattern, allowing us to predict specific ex- and record dates with great confidence.

To find out more about the other unique Woodseer features please get in touch for a free trial now.

Back To Articles