Private Equity Valuation — A Quantitative Approach
Private Equity (PE) valuation traditionally rests on the three pillars of
- Peerset multiple analysis,
- Cashflow analysis and forecasts (usually using the Discounted Cashflow Valuation model), and
- Some form of asset valuation or coverage ratio analysis.
While these are fairly robust and have done reasonably well over the years, each of these approaches has various issues, not the least being an inability to define and analyse strength of the business model and resilience of the company to external shocks — both of which have become critically important metrics in the strange times we live in.
The other issue we PE practitioners face is limited manpower, especially in small to midsize funds. DCF in particular is very time intensive and is usually only conducted once a deal has progressed a fair bit. Ideally, an efficient valuation tool can be automated to a large extent so that it also acts as a filtering tool at early stages of deal analysis.
Even for small funds, it’s not unusual to see over a 1,000 deals a year and with teams of 2–5 investment professionals, allocating time efficiently is an essential ingredient in achieving success as a PE firm.
What unfortunately tends to happen is that PE managers use heuristics, or mental shortcuts to screen firms. We tend to absorb what we can over a few quick hours of study and then arrive at a ‘gut feel’ based on perhaps previous experiences with similar firms or our read of the main guy or lady at the helm of affairs.
I’m not suggesting this is wrong, indeed PE is fundamentally a people business but when one doesn’t have the guardrails of rigorous, quantitative analysis, one is apt to slip and fall, that is to say, make errors in judgement — either type a or b (passing a good deal, accepting a bad deal).
I believe a more quantitative, rigorous approach can help enhance the PE process. The idea here is enhance, not replace. To my mind, these tools are aids to PE practitioners to help them make better decisions. Further, these tools are automation-friendly so theoretically all it should take is a click of a button to arrive at a score which can then be broken into constituent factors to analyse specific areas of weakness in the target’s business model.
I draw on the work of Wesley Gray and Tobias Carlisle as outlined in their excellent book: Quantitative Value: A Practitioner’s Guide To Automating Intelligent Investment and Eliminating Behavioral Errors. I highly recommend giving this book a read even if you don’t intend on applying their framework. It’s an eye opening experience and a very refreshing twist on the usual valuation approach.
Thanks are also due to Wesley Barnes for his help and for being a great sounding board while I was constructing this. Also thanks to him for pointing me to the afore-mentioned book in the first place!
Value Investing Concepts
The key issue for a PE manager is how do you find a ‘wonderful company’? Secondary to that is how do you avoid passing on what may seem like a bad deal but actually is a great business?
Of course the usual PE principles apply but I outline below an automation-enabled, quantitative approach to help filter even early pipeline deals.
Let’s consider two fundamental pillars of the value investing approach.
1. Franchise Power: a company has Franchise Power (FP) if it has a sustainable, competitive advantage through which it can demand pricing power and thus generate high returns on capital.
2. Financial Strength: Financial Strength (FS) is drawn from current profitability, stability and recent operational improvements. This analysis helps identify weak companies.
I won’t get into the specifics of how the formula for FP and FS is arrived at (if interested, please read the linked book!) but in short:
FP is computed as the average ranking of each company based on its ranking in the universe for:
- Return on Assets
- Return on Capital
- Free Cashflow/ Assets
- Margin Growth and Stability (computed using standard deviation and geometric mean)
FS is computed as the average ranking for each company based on whether or not it showed improving trends for
- Return on Assets
- Free Cashflow/ Assets
- Cash Accrual Ratio
- Leverage Ratio
- Liquidity Ratio
- Sales/ Assets
The final ‘Quality’ Score is then computed as
Quality = 0.5*FS + 0.5*FP
While testing the algorithm, the key issues are:
a. Data availability for the companies being analysed, and
b. Choosing the right ‘target variable’
Data for my set of companies was provided by a kind friend for which I am grateful. I’ll discuss below what set of companies I used for this analysis.
For the computation of FS and FP, I used 8 or 9 years of data as available (starting FY2010) so as to capture one full business/ economic cycle.
The target variable should be something which measures ‘success’ as an equity investor. Looking at this with a PE lens, one is tempted to scour databases of PE deals but it is next to impossible to find a comprehensive dataset. Nearly all of the commercial datasets available have serious issues with availability of numbers — especially on entry/ exit valuation. As a proxy, I decided to look at the listed space and of course, there we have share price. Yes, price has a number of other factors playing into it but at least relative to other companies and relative to the market this is an adequate measure of ‘success’.
I also used share prices for my dataset from 2018 to 2021 as my target variable (normalised to starting date). This way, I can check if the algorithm score correlates with share price performance in the ‘future’ (future from the perspective of the algorithm — so if the financial data is from 2010 to 2018 (March ending), I look at share prices from 2018 to 2021.
I considered 3 countries for this analysis:
a. India, since it has been my main region of work experience,
b. Indonesia, as another large developing market and a key component of ASEAN, also a significant area of work experience for me, and
c. USA as a control group
For India I chose the top 1,000 companies by market cap, for Indonesia the top 300, and for the US, the S&P 500.
I ran the analysis for each country as above — the final score derived from the algorithm was used to split the dataset into deciles. I then compiled the top 3 (label: ‘highend’) and the bottom 3 deciles (label: ‘lowend’) into separate groups and compared both with the entire dataset on stock price performance for the 3 year period immediately post the end date of the financial data used in the analysis.
For India, the algorithm did exceedingly well in segregating ‘good’ companies from ‘bad.
For Indonesia, the results aren’t quite as clean due to some outlier stocks but the trend does hold up.
For the US (control group) the analysis holds up exceedingly well too.
Delving into the Details
I picked up one company, (‘Company 1’) which didn’t do well — it ranked in the lowest decile and I wanted to know why.
Since Quality is composed of FP and FS, the first level analysis is to check scores for both metrics.
On FP, we see that Company 1 didn’t do that badly. In fact, comparing Company 1 to the other deciles, it actually ranks similar to the median value for all companies in decile 4 (p0.4 on the x-axis).
As seen below, it is FS that is impacting Company 1.
Further, I then examined Company 1 for the constituents of FS.
Looking at this, it is clear that it’s really only return on capital and return on assets that is pulling down this firm.
And so, not only does the process score the firms, it also manages to drill down to where the problem is relative to the universe. BTW, this particular firm had an issue with ballooning capex and debt over the time period considered but management had already announced measures to fix that (including through the sale of a few non core assets) and the firm is doing well today.
A PE manager evaluating this company using this process would be alerted to this issue and would likely pick up on the management statement and could then take a call on whether this firm merited more study or not.
Applications to PE — Other Considerations
I think the number one question on folks’ minds would be how to use this for an unlisted deal. Well, firstly post signing a non-disclosure agreement, a PE firm should have access to the data needed for this analysis. If you don’t, I’d suggest running away!
As with every deal, a PE manager will need to diligence data received (given that data from unlisted companies is usually not as ‘clean’ as for listeds). The power of this tool is to quickly crunch the numbers once available and then direct the PE manager to specific areas of concern for further analysis and diligence.
Secondly, the listed universe where data is widely available still serves as an effective benchmark. Once I know where the unlisted target company sits in the universe (that is to saw which decile it fits into) I have an idea of what sort of quality this opportunity represents.
Areas of Further Study
Further work is needed to refine the analysis:
- Metrics such as EBIT, ROCE and Gross Profit may not hold the same relative importance across industries — possibly can use machine learning to determine which metrics have the strongest relationship with the target variable per industry. Once this is known, it is relatively simple to tweak the algorithm for each industry
- Further explore private equity investee companies and corroborate the findings above for those companies. Given that data availability is an issue for those firms, it’s probably best to look at listed investee companies — this is the path I intend to take.
- Build this into a more robust monitoring tool — The break out of the scores into FP and FS can be used to identify red flags in a portfolio company and can help focus the investment manager’s time and attention.
- Wesley and Tobias in their book also talk about fraud detection algorithms — I intend to code for those and examine applicability for PE. Given developing markets in Asia and their various, ahem, idiosyncrasies, I’d be delighted to have some sort of tool to check for this.
The exercise as above shows good results in classifying companies. More work needs to be done to refine this for various industries and to evaluate how well this has performed with PE deals — I shall be working on this over the next few weeks.
In any case, this is already an effective screening tool and once the additional refinements are in place, it should hopefully also work as a monitoring tool.
The future of automation and quantitative analysis in PE is very bright!
There’s much more to be explored in how data science can be applied to generate alpha in PE and I shall discuss that in future articles.