To those of us who rank dividend growth high on the list of metrics for picking and choosing equities to stuff our portfolios with, it is sad to see the lack of coverage this subject has in most literature. This article, the first of a three-part series, presents an improved method of characterizing dividend growth, rather than resorting to three- and five-year averages.
There are two types of dividend growth: one is exponential growth -- or compounding, if you prefer -- and the other is linear growth. Exponential growth has a constant percentage dividend growth (give or take) each year. It is exemplified by Procter & Gamble (NYSE:PG), a company in the consumer staples industry and typical of the low-yield/high-dividend growth group. Linear dividend growth has a constant amount (in dollars) of dividend growth each year. Enterprise Products Partners (NYSE:EPD), as a case in point in the gas/oil pipeline business, is an MLP (master limited partnership) in the high-yield/low-dividend growth field. Real estate investment trusts (REITs) also have linear dividend growth.
The figure below depicts these characteristics. Dividend data (in dollars per year) were taken from David Fish's 2011 end-of-year Dividend Champions/Challengers/Contenders lists for the years 1999-2011.
Click to enlarge images.
The light blue and green curves are "best fit" linear and exponential calculations, which will be explained later.
If you think EPD is a better stock to own, consider what happens as more time is added to the picture. The chart below is 25 years' worth -- 13 real, 12 projected. Stocks like EPD are better short term for the yield, but ones like PG pay off in the long run if you buy and hold.
The "problem," in my opinion, is that basically the only metrics given for dividend growth are three- and five-year percent averages. These fail to tell a true story in so many cases. Given that in linear dividend growth cases, percentage dividend growth decreases every year, these metrics are misleading at best. Even in the exponential dividend growth case, caution is needed as will be seen by examining Illinois Tool Works (NYSE:ITW). For ITW, three- and five-year dividend growth averages are 6.3% and 14.9%, respectively. However, two-year averages for 2006-07 and 2009-10 are 24.8% and 5.1%, respectively. When the dot-com bubble burst, the average for 2002-03 was 4.5% . ITW is a cyclical stock and, as such, has lower dividend growth in recessions but makes it up nicely in economy growth years. The figure below shows ITW dividend history from 1999-2011.
The slope of the curve is dividend growth, easily seen going from a low value to one much higher, then back down. The low dividend growth periods are the dot-com burst and the financial crisis.
The red curve represents calculated values using calculations from a process known as non-linear regression. It finds a "best fit" to an exponential such that there are a least a number of squares between the curves. In this case, it depicts an exponential growth of 14.1%. In my opinion, this is a superior metric for dividend growth in that it is representative of the total data. (The appendix at the end of this article details the math.)
It is difficult to collect a deep history of dividends. It is also useful to compare different companies in the same time period. An additional factor is to have a long enough time period to show a reasonable sample of performance. For these reasons, an eight-year collection of yearly dividends is selected, giving seven years of dividend growth. This is also about equal to the average business cycle, so a reasonable history should be obtained.
As mentioned above, I plan to write a series consisting of three articles detailing a method for determining dividend growth in the general sense. This article addresses exponential (compounding) dividend growth, which will account for the majority of stocks. The second article will cover linear dividend growth, which covers MLPs, REITs, and other higher-yielding equities (including utilities and telecommunications). The third article will address ETFs, which distribute dividends from a collection of stocks (less fees). This area is unique because the population of equities contributing to distributions varies over time due to portfolio turnover.
The table below shows dividend growth as measured by a variety of methods. Three- and five-year averages, the normal metrics, are the average of single-year dividend growth expressed as a percentage. N-LR utilizes the method described in the appendix. This method calculates a dividend growth curve that attempts a "best fit" to the stock dividend data where the area separating the two curves is equally split on one side and the other. All calculations were made using dividend data from David Fish's CCC spreadsheets, end-of-year 2011. All data in the table are calculated in percents (%).
The companies analyzed are as follows: Illinois Tool Works, Target (NYSE:TGT), Air Products & Chemicals (NYSE:APD), Medtronic (NYSE:MDT), Automatic Data Processing (NASDAQ:ADP), Abbott Labs (NYSE:ABT), Becton, Dickinson & Co. (NYSE:BDX), Genuine Parts Co. (NYSE:GPC), Procter & Gamble, and Sysco (NYSE:SYY).
Stock Symbol | 3yr dg | 5yr dg | 7yr dg | Non-linear Regression dg (8 yrs) |
ITW | 6.3 | 14.9 | 16.0 | 16.6 |
TGT | 22.6 | 20.3 | 20.5 | 19.2 |
APD | 9.6 | 10.8 | 11.6 | 10.6 |
MDT | 14.6 | 18.2 | 17.2 | 18.5 |
ADP | 7.6 | 14.6 | 14.7 | 16.1 |
ABT | 13.0 | 10.5 | 9.2 | 9.3 |
BDX | 12.9 | 13.8 | 15.5 | 15.5 |
GPC | 5.5 | 6.4 | 6.1 | 6.0 |
PG | 9.9 | 11.2 | 11.2 | 10.4 |
SYY | 5.8 | 9.0 | 10.5 | 9.0 |
The first two columns are the standards, three- and five-year averages. The seven-year average calculation helps clarify the situation a little, but there are still deviations from the real picture. The problem is that averages inaccurately reflect the effects of the economy and will vary up and down accordingly as the updated yearly calculations progress through a downturn. Non-linear regression calculates all data equally and presents a composite picture.
A case in point is ADP, where the seven-year average is 14.7% and the non-linear regression figure is 16.1%. These are shown in the graph below. You could argue (which I don't) that the former is more accurate due to the low dividend growth in the last two years. Or you could argue (as I do) that N-LR is more accurate. ADP, like ITW, is a cyclical stock and the recovery hasn't happened (yet) with ADP.
It is better to expect (and look for) an increase in the dividend when the economy recovers than to be lulled into inaction by thinking everything is OK. In either case, we are talking about good returns.
So why is all this important, nitpicking over a percent or two in dividend growth? I am looking for a stable dividend growth characteristic for stocks that defines this important metric, one that represents the stock in all phases of a business cycle. I use this in a figure of merit, as described in "Dividend Growth As A Figure of Merit." Also, dividend growth metrics play an important part in my retirement portfolio distribution scheme as described in "Life Cycle Of Retirement Portfolios, Part 1."
A figure of merit for stocks with exponential dividend growth is "yield times dividend growth." Figure of Merits for the stocks listed in the table are (using Fish's CCC end-of-year 2011 data for yield and non-linear regression dividend growth from the table) as follows:
- IWT (51)
- ADP (47)
- MDT (47)
- TGT (45)
- BDX (37)
- SYY (33)
- PG (33)
- ABT (32)
- APD (29)
- GPC (18)
It is my intention to use this method for calculating a composite dividend growth for all stocks in my portfolio. I will compare it with updated dividend growth data each year, but I won't recalculate it unless it obviously no longer represents the stock.
Appendix
The method used in determining non-linear regression dividend growth used in this article requires the use of a spreadsheet. The following outlines the general procedure:
- Input yearly dividends in a spreadsheet. Take the logarithm [ln(div)]. It is better to arrange these in columns, dividend, year (starting with one (1)), ln(div).
- Copy the two columns, year and ln(div). Go to xuru.org. Paste data in the box. Right click to paste.
- Click on "calculate." Copy the numbers "a" and "b" from the equation y = ax+b.
- In the spreadsheet, calculate the anti-logarithms by the formulas "exp(a)" and "exp(b)." The first formula, exp(a) is (1+dividend growth). The second is the "y-intercept" or the dividend for year zero.
- Calculate yearly dividends in N-LR by the formula N-LR = "y-intercept" * (1+dg)^T, where "T" is the year and "dg" stands for dividend growth. If you arrange these in the column before the dividend data, it is easy to graph the two columns.
Example
N-LR ADP | Dividends for ADP | Year | Ln(div) |
1.1605 | < exp(a) | a = 0.14889 | |
0.4902 | < exp(b) | b = -0.71303 | |
0.569 | 0.56 | 1 | -0.5998 |
0.660 | 0.62 | 2 | -0.4780 |
0.766 | 0.74 | 3 | -0.3011 |
0.889 | 0.92 | 4 | -0.0834 |
1.032 | 1.16 | 5 | 0.1484 |
1.198 | 1.32 | 6 | 0.2776 |
1.390 | 1.36 | 7 | 0.3075 |
1.613 | 1.44 | 8 | 0.3646 |
What we are doing here is making the yearly dividend curve linear by taking its logarithm, then doing a normal linear regression. A linear regression finds a "best fit" of the data to a straight line; in essence, there are equal areas between the two curves. Doing an anti-logarithm brings the numbers back into the real world with key coefficients to plot results.
A logarithm calculator can be found here. In the example, I used the base "e," but used a spreadsheet to find the logarithm values. Any logarithm base works as long as you use the proper anti-logarithm.