Multiple Regression EV/EBITDA with various
multiples:
Factors used: EV/EBITDA was regressed
with Margin Leverage, Growth EBITDA,Growth Revenue, Return on Assets,
Capex/CFO, EPS Growth, STD stock, Cash/Assets, VIX, S&P P/E, GDP. Sector
wise accumulation.
How To Select Based On P Values?
There are various strategies to get the right
result, which are discussed in other parts.A general algorithm or pseco code
can be built up on that so that we select the best case. Some signs cannot be negative
which we know by intuition and other have to have a good p value say around
30-40% and should not impact R-Sq-adjusted much. Keeping all these in mind we
can have backward and forward selection.
Start:
Constrain on Sign – Select forward – Select Backward
– Eliminate from residuals corr/ P Values/ General Correlations – Again Observe
Rsq-adj for major change. Keeping don’t to get the best people.
Issues:
The sector shows peculiar behavior and the
company might fall into different sectors where we cannot map each sector to
the company. The macro factor like GDP, VIX, S&P, and
Unemployment seems to have less effect on the multiple which rallies due to its
own reason. If we think that we can find directions than it again becomes tough
for the reason that when EBITDA falls EV also falls so ultimately what we
get is a proxy for growth or how much we think that the pricing of high
will be in spite of low EBITDA.
On the other hand if we take intricate
parameters than we get explanation of what each component brings in the
valuations or the growth proxy. The basic notion should be that capex should
drive growth up but there may be other factors explaining, example excess cash
that will bring growth and returns down.
When we start creating excel we need to make
sure most of the things that are simple and intuitive and hence the
regression process should run in the most simple and automated way.
Another methodology suggests that we can take
important factors from various elements like leverage, growth, profitability,
macro and explain how each factor is contributing to the multiple. Again we are
searching for a balance between the statistical results and the business acumen
to get the best explanation on how the multiple is behaving and what would
happen if we tweak any parameters for the business. We would take the system
wide data in our model from each set of companies’ belonging to the same group.
Creation of the group depends on the sector, growth phase, and other
parameters.
For example if a company has a multiple of 12X
then its multiple can be explained in terms of growth, profitability, return
methodology, macro, risks (internal external). If the direction becomes
irrelevant only then we should remove the factor from the respective group. A
max and min could be put so that the explanation becomes better. If we look at
it there are inherent risk and growth in companies of different size as well as
different sectors. Directional analysis: beta, volatility, S&P risk etc
should always reduce the multiple and not positive.
Pseduo
code for performing the regression:
- Select all parameters and group them into different heads.
- Put a maz min of contributions you can take from each group.
- Make a sign constraint on the factors.
- Macro should be less than 25%.
While simulating this on excel, we have used
Linest function but there are some issues like reversal of factors and other
data handling things that makes it tough……………..
Free resources from Yahoo Finance.
over hundreds of factors? Which matters the
most which matters the least, which factors have been used most of the times.
Intrinsic vs external factors and how to avoid
patches of loop and make a scalable model in excel.
Pasting algorithms and calculating residuals
is something which becomes imp, similarity and other check. How does bounding
things to max or removing outliers work in robust regression in MATLAB? can the
same be applied in VBA manually?
Why multiple analysis is imp? it tells u how
the pricing works and which factor should be worked upon, leverage may play
negative or positive factor, ideally it plays a negative as it increasing he
risk, if stat tell the reverse do we a model that sits leverage increasing the
returns?
Stat vs theory: many negative multiples what
to do about htreml.
different companies and multiple of the
S&P how and where they last?
Factors that can used and cautions [v4]
How to regress EV/EBITDA multiple?
For sectors where margin defines the valuation
we might use margins, these sector are high margin sectors like internet,
others where margins are high.
- Gross Margin
- EBITDA/Sales
Return method might affect the multiple
depending on importance given by investors to the return quanta and
methodology.
- ROIC
- EVA
- ROE
- Cash based margins can be used instead of normal margin things.
- CFO/Sales
- FCF/Sales
- Dividend Payout Ratio
- Div Yield
- YoY DPS Growth
Growth is one thing that factors in the price
in some sectors where growth sensitivities are having greater impact. Risk
effects all sectors but different sectors have different beta based on how they
would be performing to macroeconomics sensitives.
- 1yr Beta.
- 2yr Beta.
- 5yr Beta.
- 1yr Share Price Volatility.
- 5yr Share Price Volatility.
- VIX.
- Debt/Total Capital.
- Debt/Capital Debt (YoY Change).
- Adj. Debt/Total Capital.
- 10yr Treasury Yield.
- 3mo LIBOR.
- Real GDP Growth.
- Unemployment Rate.
- S&P 500 P/E.
- Scale (log Sales, this is not a ratio).
- Square elements that you think could be very sensitive.
- Take log of elements you think can drag things around.
Careful on EV or EBITDA things as this might
skew things: Debt/Total capital is less tricky to use than D/EBTIDA because D
is on both numerator and denominator of D/Total cap where for D/EBTIDA has a
debt portion if increasing causes multiple to go up.
Past year or next year? The data fetching is a
problem here in terms of how the market would react and feedback mechanism.
IV Considerations:
Things that we give greater priority than R
square:
- R-square with spurious correlations.
- With negative coeff (not making sense).
- Where some elements explain all and very few left for others (client) {group vs client).
LTM is better option than 3 years. Values are
at point in time; taking them average of quarters would create smoothing of
data points?
We do a similarity check to remove companies
that has data that creates trouble for us.
I think we use both the combined as well as
the component versions? Between is the FGR definition the same as given above?
Arithmetic average of (NTM rev growth, NTM ebitda growth, “long term” EPS
growth), where the last is whatever the analyst’s report – typically 3y or 5y
outlook.
10Y UST stands as the standard one, other
options are 1 yrs 5 yrs and 30 yrs.
I guess we are using the change and not the
absolute; change should be used not absolute as change is something that makes
things move up and down. This notion is important because of the change in
multiple could be due to lag or the expected growth in the future. That is the
reason use NTM vs LTM. This can be understood as feedback or growth expectation
driving the multiple. Similar analysis has been done in AR1.
Which part plays the biggest contribution on EV/EBITDA multiple in MLR?
Biggest contribution is of estimated
growth; otherwise - acq - biased - 2008 market acquisition.
Other than the companies internal numbers - if
we take all factors than S&P dries the multiple.
Is the history irrelevant?
What explains D/EBITDA?
If we use correlation analysis for
point in time - as a supplement of our regression analysis then we might
think that growth and sales should explain it.
Each sector has different dynamics to see the contrast we should look at following sectors - defense companies - mature - chg
- mature - Biotech - research shoot up - biotech -
energy - industrial EV/sales.
Intrinsic for MC sim
- Sales
- Multiple
- Margin
- EV-EBITDA
Market numbers:
- Interest rate
- S&P
- Libor
- Steepness
No comments:
Post a Comment