Tom's Life Bits

There are stats that I calculate that expose more of my thinking than I want to have exposed - so I do not show them often. At year end - and being almost fully invested - I had decided to share those stats. I really hope that most of you do not go to the effort to understand them - and that most of you believe this is a bunch of voo-doo math that will never work in the real world. That is partially a correct impression - because I will show that the stats are only about 70% predictive and 30% in error.

Given that the statistical analysis is NOT the end of my investigations - and not the sole determinant of my investing decisions, I can live that with level of error. The statistical analysis is just the beginning - it highlights what MLPs I want to investigate further.

And the first step in generating predictive numbers is to generate good forward CAGR [compound annual growth rate] numbers. The first spreadsheet shows where I get those numbers - and why the numbers I use differ from the CAGR estimates the one can find at Yahoo! Those numbers are in the next message.

MLP CAGRs - Why Consensus Estimates Are Not Used

The dominant factor in assigning amended CAGRs is the distribution coverage ratio, with forecast DCF growth second and recent distribution growth third. My CAGRs are close to the analyst estimates - with only a few exceptions.

I had thought about leaving out part five - it being 'the truth' - and some folks can't handle the truth [grin]. But what the heck - if you can't handle it, then you can just ignore the message. Part five is a short discussion on why I justify placing a large part [but not the WHOLE part] of my investing decision based on numbers.

Having a numeric representation of the forward projections works for me - even knowing up front that there will be errors in those numbers. If making decisions with my gut worked for me - then I would use that. But the 'numbers' method not only works in netting better returns - it works a WHOLE LOT BETTER on the psychological level.

I can accept my numbers being right only 65% of the time and combining metrics of 65% correctness to raise the level of accuracy in their predictions to around 70%. If my gut was only 70% accurate [and if I were using my gut to make decisions], then I would be questioning myself left and right - wondering why I was such a failure - wondering how I could be so smart in 2006 [where investments in ETP and BWP led my portfolio to soundly beat the sector average] and so dumb in 2007 [where my investments in ETP, BWP and PAA caused my MLP portfolio to slightly under-perform the sector average].

I can also work hard on getting the numbers right - adding new metrics - changing the weighting of different variables. Adding SSNOI and rent spreads should make my metrics on REITs better - or more predictive. Tracking changes in book values should make my bank metrics more predictive. Testing if the MLP metrics are more predictive using current year projections are next year projections - that improves the model.

It is hard to add something to your gut gut without adding weight. [grin] Can you listen to EVEN MORE conference calls? Examples: PAA promises 10% distribution growth, but ETP fails to promise anything. Does that make PAA the better choice? The management of ETP is a bunch of lying jerks [they have told me to call back from a conference call to get data - then stonewalled on giving the expected data], but PAA's management seems to be straight shooters. Does that make PAA the better choice?

Going with the gut puts intangible questions like THAT in your mind. I believe you can spin those mental wheels forever and get nowhere. Or maybe I was never taught WHAT to listen for - so I was listening and nitpicking all the wrong things.

At least with numbers - I know to what degree I am changing, if things go wrong. I can add weight to the dividend discount model and cut the weighting to the current price. I can add to the price at a logical P/E ratio and cut the weight to the price at a logical Price/DCF model. I can add target prices to the mix. I can back-test the data.

One can not do similar back-testing on the weighting of 'promised distribution growth vs. the promising of nothing'. And if ETP outperforms PAA - then maybe ETP just under-promised and out delivered, turning a negative into a positive. So my mind must have been fooled hearing a negative in the first place. Right? Wrong? Heck if I know!

So that is my thinking - and why I like numeric decisions better than gut decisions. That is why I tend to focus on the numbers. A focus on the numbers is sometimes a focus on the facts. And even when the numbers are not facts [a lot of numbers I use are only opinions], at least they are quantified opinions. And one can do something with a quantity. I can believe that the management at ETP is a bunch of @ssholes - but what does one 'do' with such an opinion?

So I let the numbers play a large part of my decision process. And this could also be the reason why I still own units in ETP - its the numbers. And it is the reluctance to pay a lot of deferred taxes.

Enchilada part seven - the very long explanation

- a fictionalized narrative and an introduction to rocket science

The other day my neighbor dropped by to ask questions about some of the spreadsheets I had posted on the IV board, and on other spreadsheets done by the brokerages or done by me and not recently shared with the good folks at IV. And he was complaining that 'there are just too many dang numbers. There is [1] EPS - and [2] the trend in EPS growth. There is [3] DCF [distributable cash flow] estimates and [4] DCF growth trends. There are [5] CAGRs - and some of those CAGRs look kinda flakey. There are [6] distributions and [7] trends in distribution growth.“ [And we will temporarily ignore some of the other numbers he listed - like ratings and target prices.] “It's sort of like juggling every time I have to make an investment decision. And just like juggling, I can handle two balls and sometimes three - but when you get to that fourth ball - the balls (or in this case the spreadsheets) start going every which way - and I just get flustered. And that is just with analyzing ONE MLP. When you start to compare two MLP’s against each other - that adds up to fourteen dang balls in the air at once. There has got to be a simpler way!”

I looked my neighbor in the eyes and smiled - and told him “Boy, I've got a spreadsheet for you!” What if I could boil down growth into just one number - a number that combines the trends in EPS, DCF and distribution growth? You already know that number - its CAGR. But we are NOT going to use the analyst's CAGR - because you are right - some of their numbers ARE flakey. We are going to start with their number - then look at the trends in EPS, DCF and distribution growth, then come up with a number for CAGR that looks right to us. If you have a good CAGR, then you can use that number in combination with its distribution amount - and via the dividend discount model [he was a bit foggy on the DDM - so I told him to just Google it later] and come up with a price estimate - - - a price towards which the stock - all other things being equal - will be headed.

If you have a forward EPS estimate - then you can use that number to generate a modeled price. And if you have a forward DCF estimate - then you can use that number to generate a modeled price.

But why would you want to do that!?!? Are you just creating more balls to juggle?

No I said. We would be creating three new and different balls - but they are 'combo' balls - balls that contain the seven elements you listed above. So instead of having seven balls to juggle, you now have only three - which makes juggling them a whole lot easier.

Let's start with EPS, and take three fictional stocks, all with an EPS estimate of $2.00 per unit. But one has a historical trend of growth of 4%, one of 8%, and one of 10%. Since they all have the same 2008 EPS estimate - should they all be priced the same?

“Of course not” he said. If they were all priced the same, then people would sell the 4% growing stock, driving down its price over time, and buy the 10% grower, driving up its price. They could start out at the same price, but they would not stay that way for long. I then asked if all three stocks began at the same price and the same yield - then what would happen to the yield as the prices changed. He said the yield on the slow growth stock would rise, and the yield on the high growth stock would fall. My neighbor was quickly catching on.

OK, right now, the P/E ratios range from 14 to 25. These are real values that real people are placing on real stocks and paying real money. While the average P/E is around 20 - there is no bell curve distribution of P/Es. And that is a bit strange - because there is a bell curve distribution of CAGRs - with the average MLP having an 8 and a range from 4 to 11. So without doing any real math - lets make a guesstimate of what the P/Es for these three stocks should be.

Well, that simple he said. The 8% grower is an average grower - it should have an average P/E - which is 20. The 4% grower is at the bottom of the CAGR list, so it should have a P/E of 14. The 10% grower is close to but not at the top of the CAGR list - so it should have a P/E of . . . say around 22. I said that sound pretty logical to me. Now let's make a mathematical formula that uses EPS and CAGRs to consistently give a fair ratio to each stock given its CAGR attribute.

The sum of 11 plus the CAGR would generate a P/E range of 15 to 22. It would generate too high a price for the lower CAGR stocks and perhaps too low a price for the high CAGR stocks. But folks are yield hogs. So if I am going to generate a modeled price, I want to take that fact into account. They will tend to over-value the high yields and under-value the low yielder's. So let's cheat the high yields a bit in this model, but make it up some in the next model. But, he said, “I still do not know why the heck we are spending time making model prices!”

But, I said, “we are not really making model prices - we are making combo balls.” We are going to make combo balls using CAGRs and DCFs - but before we do, let’s look at some real world examples - all with stocks that have close to $2.00/unit EPS estimates in 2008. Let's see how they fit with the modeled prices - and let's see if the modeled prices might make more sense than the actual prices.

BWP's 2008 EPS estimate is $1.92. BWP closed on 1-04 at $31.00 and has a CAGR estimate of 9.0%.
DPM's 2008 EPS estimate is $1.95. DPM closed on 1-04 at $42.32 and has a CAGR estimate of 10.0%.
HLND's 2008 EPS estimate is $2.12. HLND closed on 1-04 at $48.70 and has a CAGR estimate of 10.5%.
KMP's 2008 EPS estimate is $2.05. KMP closed on 1-04 at $55.21 and has a CAGR estimate of 8.0%.
TLP's 2008 EPS estimate is $1.89. TLP closed on 1-04 at $30.50 and has a CAGR estimate of 10.2%.
TPP's 2008 EPS estimate is $2.03. TPP closed on 1-04 at $39.22 and has a CAGR estimate of 4.0%.

All five have EPS estimates within 6% of $2.00 - but with prices that range from a low of $30.50 to $55.21. That is a low range of variation from the $2.00 EPS estimate - and a very wide range of variation in price. Let's quick do the calculation of a modeled price at a logical P/E ratio and see a different look at the price variations.

BWP's modeled price is $40.32 vs. a closing on 1-04 of $31.00 - a 40.32% discount.
DPM's modeled price is $42.90 vs. a closing on 1-04 of $42.32 - a 1.37% discount.
HLND's modeled price is $47.70 vs. a closing on 1-04 of $48.70 - a 2.05% premium.
KMP's modeled price is $41.00 vs. a closing on 1-04 of $55.21 - a 25.74% premium.
TLP's modeled price is $41.96 vs. a closing on 1-04 of $30.50 - a 37.57% discount.
TPP's modeled price is $32.48 vs. a closing on 1-04 of $39.22 - a 17.19% premium.

My neighbor said “Hmmm . . so what?” I told him that we are just getting started. EPS is only one metric or model - and we need more than that to judge the fairness or appropriateness of the current prices

Using the same process as used with EPS, we created combo balls of CAGRs and DCFs. That produced the following data:

BWP's modeled price is $33.75 vs. a closing on 1-04 of $31.00 - an 8.87% discount.
DPM's modeled price is $45.72 vs. a closing on 1-04 of $42.32 - an 8.04% discount.
HLND's modeled price is $67.56 vs. a closing on 1-04 of $48.70 - a 38.72% discount.
KMP's modeled price is $50.41 vs. a closing on 1-04 of $55.21 - an 8.69% premium.
TLP's modeled price is $42.48 vs. a closing on 1-04 of $30.50 - a 39.30% discount.
TPP's modeled price is $28.75 vs. a closing on 1-04 of $39.22 - a 26.70% premium.

Through a different process, we created via the DDM [you will Google that later if you need to] a combo ball of the current distribution and the CAGR. So, my neighbor asked, “what do we do with these model prices?” “We test them” I said. What good are model prices if they are not predictive! “You mean they may NOT be predictive?” he said. I explained that they are far from being a sure thing. The EPS estimates will go up and down during the year - and the same for the DCF estimates. This will mess up or change the model prices. This is going to happen somewhat randomly. Will these random changes make the projections in the model prices garbage - or will they still be useable? And to test the models, we use historical data.

We looked the EPS combo balls first, using the EPS estimates that existed in January of 2007. It would be cheating using the current estimates.

The following companies had logical P/E valuations that were more than 10% above the 2007 beginning price:
CPNO, ETP, GEL, MMP, MWE, TCLP and WPZ. Their mean price gain for the year is 9.95%. Their mean total return for the year is 15.99% - and 4 of the 7 beat the sector median yearly price gain [3.47%].

The following companies had logical P/E valuations of less than 10% above the 2007 beginning price : APL, BPL, BWP, DEP, DPM, EEP, EPD, EROC, HEP, HLND, KMP, NGLS, NS, OKS, PAA, RGNC, SXL, TLP, TPP, EXLP, XTEX, KSP, MMLP, TGP and USS. Their mean price gain for the year is 1.66%. Their mean total return for the year is 8.2% - and 9 of the 25 beat the sector median yearly price gain.

My neighbor said lets looks at the group with the 10% discount [CPNO, ETP, GEL, MMP, MWE, TCLP and WPZ] - which of those did not beat sector average and why did they not beat the average - if the model is so predictable? OK. The three that had less than sector average years were ETP, TCLP and WPZ. ETP got in trouble with FERC - hurting its forward outlook. WPZ had a falling DCF estimate. And TCLP, despite having 10% distribution growth in 2007, has only had 3.74% average annual distribution growth since 2001. So it is currently a mid-growth MLP that the market prices as a slow growth MLP due to that history. TCLP ended 2007 yielding 7.29% in a sector where the average yield was 6.37%.

My neighbor asked: “Are there stories like that to explain why some of the MLPs with lower growth in their modeled prices had better than average years?” Yes. DPM had a lower discount because it began the year at a low yield anticipating its forward growth. But during the year, the DCF estimate grew by around 20%. EPD was up 10% because their organic projects were coming on line. It was in the news - and the news was good. And stocks in the news rise more. But EPD ended the year yielding 6.15% where the sector average yield is 6.37%. Should an MLP with close to average distribution growth yield less than sector average? Well . . that is close to average. So maybe I am nitpicking on that one. I own it. It is probably slightly over-valued - but only slightly.

My neighbor asked: “So what's the point of doing the modeled prices again?” I said - it narrows your menu. If you chose to put new money into a stock in January and chose from the discounted menu - the average return would have been 15.99%. If you chose from the menu with the smaller discount, the average return from those stocks was only 8.20%. From which menu would you want to choose?

“Oh” he said - “I think I get it. Despite all the noise and the changes in EPS estimates - the model WAS still significantly predictive. Do you have data from 2006?” Yes, I said. 2006 was a very good year for MLPs. And because there were higher spreads in the total returns between the winners and the . . ah . . less winners, the model was even more predictive.

The following companies had logical P/E valuations that were more than 10% above the 2006 beginning price: BPL, BWP, CPNO, ETP, HLND, MMP, OKS, PAA, SXL, TCLP and MMLP. Their mean price gain for the year is 35.86%. Their mean total return for the year is 43.28% - and 7 of the 11 beat the sector average yearly price gain [23.14%].

The following companies had logical P/E valuations of less than 10% above the 2006 beginning price: APL, EEP, EPD, HEP, KMP, MWE, RGNC, TLP, TPP, VLI, WPZ, XTEX, KSP, TGP and USS. Their mean price gain for the year is 13.82%. Their mean total return for the year is 20.76% - and 3 of the 15 beat the sector average yearly price gain.

So choosing from the 'menu' of the highly discounted MLPs netted an average total return that was TWICE the sector average return. I would call that pretty dang predictive. He asked if I had the numbers on the DCFs. I did - and I showed them those. He asked if I had the numbers on the DDM model. I did - and I showed them those. I then explained that I also calculated a four model average - and number that combines the valuations for the EPS model, the CAGR model, the DDM model, plus the current price, and so one has a single number to compare to the current price to see if there is a discount. It is logical that if each of the predictive models is in fact predictive, then it is logical that the average of those four models would also be predictive.

Now, let's go back and look at the DCF projection for BWP, DPM, KMP, TLP and TPP once again. [Reposting that data]

BWP's modeled price is $33.75 vs. closing on 1-04 of $31.00 - an 8.87% discount.
DPM's modeled price is $45.72 vs. closing on 1-04 of $42.32 - an 8.04% discount.
HLND's modeled price is $67.56 vs. closing on 1-04 of $48.70 - a 38.72% discount.
KMP's modeled price is $50.41 vs. closing on 1-04 of $55.21 - an 8.69% premium.
TLP's modeled price is $42.48 vs. closing on 1-04 of $30.50 - a 39.30% discount.
TPP's modeled price is $28.75 vs. closing on 1-04 of $39.22 - a 26.70% premium.

My neighbor asked “so if I were buying an MLP today, the odds are saying that I am better off buying HLND and TLP at their huge valuation discounts to the current price - and I should be reluctant to buy TPP and KMP?”

I replied that I know that there is noise in the system - so ignore the under and over valuations of less than 10%. The system paints with a very broad brush - so look for the big discounts and the big premiums. The big premium from the modeled price for TPP scares me. The big discounts that one can get from buying HLND and TLP - those really look attractive. BWP and DPM are high growth MLPs - but they are close to fairly priced given what the market is willing to pay for them today. But even my modeled prices takes into account that growth always sells at a discount. The 8% of a premium for KMP is not that big of a premium for a large capped old timer with one heck of a good distribution growth history. KMP sells at a huge premium based on the EPS modeled price. But when the DCF and EPS models vary - it is usually due to the EPS estimates being wrong.

We switched our focus back to the spreadsheet for all MLPs in the sector. My neighbor asked why I just didn't buy ALL the stocks on the 'discounted' menu. I explained that I did not buy holdings of less than 1% of my portfolio. I had some MLPs that were more than that. But I also wanted to maintain my asset allocation model which kept me placing too many eggs in one basket. So the asset allocation limitations kept me from buying them all.

My neighbor said “OK, so you have only two years of data showing this cock-eyed system works . . . ”. I stopped him before he could go on. It is not just two years in one sector. I have tested this on banks and REITs. And this system is consistent - that the menu of stocks selling at a discount to logical valuations outperforms the menu of stocks that sell at a premium. It has worked for four years in REITs [I started testing them first], three years in MLPs, and two years in banks.

Over time - stocks will hover around their logical values. On any given day, the price could be set by greed and momentum - just look at the prices one could have paid for CPNO or GEL this year. Or look at tech stocks prices at the beginning of 2001. And on any given day, the prices could be set by fear, uncertainty and doubt. Valuations are logical. Valuations are based on the fundamentals. Price modeled valuations are more logical and, in one sense, are more real than real prices. That is because real prices can - and do - lose all touch with reality.

My neighbor asked why I held stocks in my portfolio like EPD, which was not on the discount menu. I explained that I did not want to churn my portfolio and sell a stock when it dropped from the discount menu. And besides, some of the stocks NOT selling at a discount still had good yields and good growth prospects.

My neighbor said “OK - enough about history - what are your spreadsheets saying to buy or not buy today, and why?” We looked at two stocks: MWE which the spreadsheets indicate is under-valued, and TPP, which the spreadsheets say is over-valued.

MWE has an analyst CAGR of 13.4%. It has averaged 128% EPS growth per year going from 2005 to its 2008 projected EPS. DCF per year growth from actual 2005 to projected 2008 is 19.96%. Actual distribution growth from 2005 to 2007 has been 17%. The 2008 distribution to DCF ratio is 135.91. [Hey - my neighbor said - you did not explain that metric to me. I told him he probably he all the explaining he could handle in one day - and we could talk about that one next week] I believe I am being hyper conservative at giving MWE a CAGR of 11%. The 2008 EPS estimate is $1.60. $1.60 times [12+11] = a modeled price of $36.80. The 2008 DCF estimate is $2.99. $2.99 times [6+(11/1.2)] = a modeled price of $45.35. MWE closed at $34.56 - meaning it sold at a 6.48% discount to its EPS modeled price and a 31.22% discount to its DCF modeled price. MWE was at a yield of 6.37% - which is right at sector average. Should a stock with the projection of significantly higher than average growth sell at only a sector average yield? And an MLP with a 6.37% yield and 11% distribution growth would average a 17.37% total return.

TPP has an analyst CAGR of 5%. It has averaged 5.56% EPS growth per year going from 2005 to its 2008 projected EPS. DCF per year growth from actual 2005 to projected 2008 is 3.60%. Actual distribution growth from 2005 to 2007 has been 1.48%. The 2008 distribution to DCF ratio is 110.79. I believe I am being liberal giving TPP a CAGR of 4.0%. The 2008 EPS estimate is $2.03. $2.03 times [12+4] = a modeled price of $32.48. The 2008 DCF estimate is $3.08. $3.08 times [6 + (4/1.2)] = $28.74. TPP closed at $39.22 on 1-04-08, meaning it sells at a premium of 17% to its modeled EPS logical value and a 26% premium to its DCF modeled value. An MLP that is yielding 7.09% and growing its distribution at 4% gives one an 11% return - which it pretty good compared to REITs and banks - so the price is not likely to fall a lot. But it does not compare well with other stocks in this sector, which I would image would result in less [significantly less] than sector average unit price growth.

Finally, as the neighbor was leaving, he asked about the predictivity of ratings.

The following companies were the highest rated MLPs at the start of 2007 - those that had ratings of less than 2.25 at the beginning of the year [a rating of 1=strong buy and 5=strong sell, so lower ratings are better ratings]: APL, BWP, DEP, DPM, ETP, EROC, GEL, HEP, HLND, MMP, MWE, NGLS, RGNC, EXLP, WPZ, TGP and USS. Their mean price gain for the year is 5.37%. Their mean total return for the year is 11.65% - and 8 of the 17 beat the sector average yearly price gain. The following companies had ratings of more than 2.25: BPL, CPNO, EEP, EPD, KMP, NS, OKS, PAA, SXL, TCLP, TLP, TPP, XTEX, KSP and MMLP. Their mean price gain for the year is 1.32%. Their mean total return for the year is 8.06% - and 5 of the 15 beat the sector average yearly price gain.

So yes, ratings are slightly predictive. But my other metrics are significantly more predictive. So why even look at something that is LESS predictive?

My neighbor started to ask another question - but I raised my hand and said ENOUGH. I have done all the explaining that I want to do for one day. Maybe you should sum it up by telling me what you learned.

The combo-balls thing - it's a bit weird, but I get it. Like a P/E or Price/DCF ratio is only half a story. Some MLPs deserve higher ratios due to their projected growth. So a metric like the discount to the price at a logical price/DCF - that tells the whole story. You know that stuff happens and EPS and DCF metrics change. But just because stuff happens, that is no reason to ignore the forecast. You have back tested the metrics, and the MLPs that sell at discounts to their logically modeled prices outperform those that sell at premiums. So if one listens to the story that the numbers are telling you, you can significantly increase the odds of having a winning portfolio.