Non-parametric concordance analysis was performed for 100 dividend-paying stocks to estimate the proportion of dividend payouts that increased (decreased) over the last 13 years. Stocks that have not paid dividends continuously for the last 13 years were analyzed going back to their earliest payout date.
The results of concordance analysis range from -1 to 1, where a -1 indicates that dividend payouts almost always decreased and 1 indicates the dividend payout increased the majority of time. A randomization technique was used to randomly select pairs of historical payouts 500 times, for which the product of sgn[payout(j)-payout(k)] x sgn[time(j)-time(k)] for the pair of payouts (j,k) and their corresponding times where summed. If time(j) was earlier than time(k), then sgn[time(j)-time(k)] was -1, whereas if time(j) was later than time(k), then sgn[time(j)-time(k)] was 1. Therefore, if a payout increased when comparing a later time with an earlier time, the product is 1, and minus 1 if the payout decreased. When the sgn() of the dividend difference and sgn() of the time difference are the same (i.e., "concordance"), their product will be 1. However, if the value of sgn() for the dividend difference and sgn() for the time difference are different ("discordance"), their product will be -1.
Once random sampling was done, the concordance estimate is determined as the sum product divided by 500 -- which essentially indicates the proportion of time the dividend payout increased or decreased. Concordance values near zero reflect little change in the dividend payout over time. The randomization approach helps minimize the bias introduced when comparing stocks with varying time periods within the last 13 years during which dividends were paid out -- as not all of the stocks considered paid dividends for the full 13 years.
The above approach to concordance is also non-parametric, since it is not based on parameters such as the mean, standard deviation, etc. Non-parametric methods also relax the normality assumption which requires use of smooth non-skewed distributions without outliers and spikes. If the distribution of payouts is curvilinear and jumpy with a rich structure of spikes and outliers, then parametric methods to determine trend such as regression or curve fitting would break-down and not reliably reveal the general trend of whether payouts increased or decreased over time.
The top 40 stocks are listed in the table (image) below, and are ranked according to a score that is equal to (%institutional ownership) x (yield) x (concordance). Thus, these stocks have the greatest history of a continual dividend increase, with high percentage of institutional ownership and intermediate yield (4-7%). As an example, the concordance value for Pitney Bowes (NYSE:PBI) is 0.88, which means that over the history of dividend payouts, PBI increased their dividend about 88% of the time. Investors with a shorter time horizon before retirement who require greater levels of yield may be interested in the stocks listed at the top of the table below.
Click to enlarge
Click to enlarge
The bottom 40 stocks are listed in the table (image) below. These stocks have the lowest score value reflecting lower concordance weighted by % institutional ownership and yield. Several of the stocks have high concordance values, suggesting a strong history of increasing their dividend, but when combined with institutional ownership and lower yield, the result is a lower score value. You should probably avoid the first 10-15 stocks in this table, but many others thereafter have lower scores because of low institutional ownership or lower yield -- but are safe to own. (Note: CH is a mutual fund).
Click to enlarge
Click to enlarge
Non-parametric concordance analysis is useful for determining the per cent of a stock's history for which the dividend increases, decreases, or remains constant. The randomization approach helps minimize bias introduced when comparing stocks with varying length of payout history. Obviously, there is no guarantee that the increases (decreases) in payouts will continue in the future. Certainly, there are other characteristics which can be used to gauge future payout growth (reduction), however, this report focuses on concordance.
The intent of this report is to reflect how a randomization approach to non-parametric concordance can help encapsulate in a single value what has happened to a stock's dividend payout over history. Non-parametric concordance can better avoid outliers and jumpy transitions of payout values over time, and when combined with randomization can help reduce the effects of varying length of payout history for multiple stocks.
Disclosure: I am long AJG, CH, CPNO, D, ETP, HCN, HIX, LGCY, LINE, MCD, MCHP, NLY, NS, PWE, SCCO, SUI, WIN.