How do YOU DCF?

Many a young financier becomes enraptured with the power of the discounted cash flow model in their third year finance classes, and student investment funds are teeming with DCF models that do three way sensitivity, have 4 growth stages, and model 20 years into the future. However quickly students feel cheated by this tool that promises to value everything with a income statement.

At first it's the baseline figures; what should the beginning change in net working capital be, what if it's been on a negative growth path for the last 3 years. Back in 2009, baselines were really ugly; what should you do with deferred tax loss, a negative NOPAT, uncertain CapEx expenditures. Then it's growth rates, which are especially tricky for growth companies. Soon you end up with a value per share for Google at $103 and you wonder where you went wrong.

I'll outline how I approach some of the many DCF decisions and then I'll talk about where the DCF really shines - when it's used backwards.

Revenue
This is the best line item to spend the majority of your time modeling because it's the hardest to fake. Managers can push and pull on SG&A, depreciation, tax expense, and really everything else before net income, but there is very little you can do (legally) to move the top line around.


I like to use one of three approaches to model revenue growth, regressions, market share & growth, and historical. Regressions are by far my favorite, especially for resource companies. How much correlation do you think there is between Nobel Energy Corp's historical revenue and the yearly price of oil multiplied by the barrels of oil produced per year? 85%.

Yes it's a very short development period, but it tells a good story, and it makes sense (when oil price drops so does revenue). The biggest plus is with regressions you are able to use industry data to forecast the company. In this Nobel Energy example I used the EIA oil forecast (see my "must have links" post) and with this I can model out 30 years if I desire. If the company you are tackling has been pretty stable for the last 10 years (no M&A or spin offs) you'd be best to create your model in a development period and then test it on recent ex ante data to see how it performs. For example create the formula on data from 2000 to 2006 and then see how it performs from 07 to 08. Or develop through 08 and test 09 to the TTM. Most models might break through the crash period, so give your power tests some slack if you use this as your test period.

If you are valuing a consumer goods or manufacturing business you will be most interested in market share and market growth. Analysts produce a lot of research on which companies are likely to gain or lose market share, and how much the entire market is expected to be worth today and tomorrow. The gathering and aggregating of this data will be much messier than resource data, but if you are able to put it all together I'd bet your model power would be far stronger than just extrapolating on historical data.

Sometimes you need to use what you have in the company, and although it hurts there are ways to make the best of it. The big first step is to dive deep into the MD&A and scour past conference calls to get a good idea on managements expectations for growth. Any negative expectations should be taken very seriously, any very positive expectation should be taken with a grain of salt.

NWC, CapEx, Dep, EBIT
So after spending all that time on modeling revenue, must we do the same for each DCF component as well. I say we do not! Whittling to our free cash flow using the formula:

[Rev-Exp=GM-Dep-Tax=EBI+Dep+Chg_NWC-CapEx=FCF] 

We are just left to consider proportions and margins before the terminal cash flow. If management gives firm CapEx projections than you can hard code those in if you feel they will happen, otherwise look closely at the CapEx proportion to revenue over a historical period that represents the future 5/10 years. Same should be done with depreciation, NWC and the profit margin to EBIT to estimate expenses. Generally if the company is doing more business all of these items should increase roughly linearly. The exception of course is if the business is shifting from one business cycle to another. If you feel the company is losing its "core competency" or competitive edge then you should make some adjustments to the EBIT margin you assume in the future. This is tricky business and you might want to look at a historical example. For instance you might look at Microsoft's margins coming into 2000, to model Google's next 10 years.

Terminal Cash Flow
This whole calculation is a headache in itself, but extremely important as it usually makes up ~2/3 of the firm value. If the firm is steady state you could probably do the terminal cash flow anywhere from 5 to 10 years out. Theory and historical data suggests that the profit margin should squeeze that the firm will increase dividend payments (or start them) so your plowback is smaller. If the firm is growth, you might want to go through the exercise of smoothing to a mature stage over a 10 to 15 year horizon. Generally this is a messy business and it is hard not to feel like you are crudely making up growth factors when it's 2021.

Making the DCF do something useful
So after playing with your DCF and making it produce a price that fits your thesis, it's hard to think that the model is anything more than a elaborate way of making up a stock price. And honestly, without rich financial data, a professionally developed model, and a team of 7am to 11pm working analysts your not going to generate alpha. You can however make the model do something useful.

By accepting Semi Strong Form Efficiency instead of WFE we can say that the stock price is the correct price. This allows us to chose a different output variable, like equity risk premium for our DCF model (hope you soft coded it, so you can use solver). The beauty of it is, if the output variable is a 'common variable' (not firm specific) we can use several different firm DCF models to narrow down on a more accurate value. It's like triangulation in mountaineering, but in finance! For instance by modeling all of the big companies in the auto sector (and yes going through the work of regressions and margin estimates for each of them) we can turn all the models around and figure out what equity holders demand as a risk premium. Of course adjustments for outliers will be needed and you may want to weight your average to focus on 'pure' peer firms in the industry. Boil this down through all the sectors and you have completed a much more robust and market sensitive estimate of E[Rm], which we can feed back into either the CAPM or Famma French Model to make passive investment decisions. The power obviously would be found in narrow sub sectors where you are likely to be have some added information, but the method is actually a widely used practice.

What actually keyed me onto this method was a paper by Aswath Damodaran. You can find the paper here, and it nicely details some of the implications of the E[Rm] DCF solution method near the last third.