Picking Comparables by Economics Instead of Sector Label

The default way to build a comparables list is to click a sector filter and take what comes out. It is fast, it is what every screen makes easy, and it produces peer groups that mix businesses with almost nothing in common financially. A sector label answers the question "what does this company sell". It does not answer "does this company convert revenue into cash the way mine does", and only the second question has anything to do with what a multiple should be.

Rebuilding a peer set on economics takes about an hour with free filings. The reason to bother is that the implied value usually moves by a lot, and it moves in a direction you can explain, which is more than can be said for the sector median.

What a sector chip is actually grouping

Take the technology bucket, since it is the worst offender and the one most retail portfolios are overweight. Inside a single technology filter you have subscription software businesses with 80 percent gross margins and capital expenditure near zero, hardware manufacturers with 35 percent margins and factories, and companies that build data centres and spend more on capital equipment in a year than a software business will spend in its life.

Those three sets should not trade at the same multiple and they do not. Averaging them produces a number that describes none of them. If your company is in the third group and you compare it to a sector median dragged upward by the first, your company will look permanently cheap, and you will keep buying it and keep being confused.

Sector classifications were not built for this. They were built to organise indices, and they group by end market because that is what index construction needs. Using them as an economic grouping is a borrowed tool applied to a job it was not designed for.

I still start at that chip row, because the Company Valuation Engine screen makes it trivial to get a filtered list with market caps and multiples attached, and the export control turns it into a spreadsheet. The chip gets me a candidate pool. The next hour is spent throwing most of the pool away on grounds the chip knows nothing about.

The three numbers that define an economic peer

Three figures, all available in the filings, sort a candidate pool better than any classification.

Gross margin structure. Not the level alone but the shape. A business at 78 percent gross margin has a fundamentally different cost base from one at 34 percent, and no amount of shared end market changes that. Pull five years so you can see whether the margin is stable or drifting, because a drifting margin tells you the competitive position is moving.

Capital intensity. Capital expenditure divided by revenue, averaged over five years. This single ratio separates businesses more cleanly than anything else on the list. Under about 5 percent of revenue you have an asset light business whose earnings approximate cash. Above 15 percent you have a business where earnings and cash flow diverge structurally and multiples on earnings mean much less.

Reinvestment rate. What share of after tax operating profit goes back into the business to sustain growth. Compute it as the sum of capital expenditure less depreciation plus the change in working capital, divided by after tax operating profit. This is the number that ties growth to cash, and it is the one nobody looks at. A company growing 15 percent while reinvesting 20 percent of profit is a fundamentally better business than one growing 15 percent while reinvesting 70 percent, and the two will never be worth the same multiple regardless of what sector they share.

Score every candidate on those three, then keep the ones that sit within a band of your subject on all three at once. Being close on two and far on the third is not close.

How far the implied multiple moves

Here is a worked example with the arithmetic laid out, so you can see the size of the effect rather than take my word for it. Suppose your subject company is capital intensive, spending around 18 percent of revenue on capital expenditure, with gross margin in the mid thirties. The sector chip returns nine candidates.

The median of all nine is 22 times. The median of the four that actually share your subject's economics, A through D, is 15 times. That is a 32 percent difference in the multiple you apply, which flows straight through to the value: a company you would have called 20 percent undervalued against the sector median is fairly valued or worse against its economic peers.

The pattern in that table is not a coincidence I constructed. Gross margin and capital intensity are strongly related to the multiple a market assigns, because both feed directly into how much of each revenue dollar reaches an owner. When you sort candidates by economics you are sorting them by the thing multiples respond to, which is why the resulting set is tighter.

Notice also what happened to the spread. The full sector set runs from 12 to 44 times, a range so wide that the median is barely meaningful. The economic subset runs 12 to 18. A tight range is the sign that you have found real comparables, and if your economic set is still spread from 12 to 40, you have not finished filtering.

An hour of work, in order

The routine, for a single name, assuming you have a screen that exports.

1. Pull the sector list and export it. Five minutes.
2. For your subject, compute the three figures from the last five years of filings. Fifteen minutes the first time, five thereafter.
3. For each candidate, compute the same three. This is the slow part. Twenty candidates at two minutes each is the bulk of the hour, and you can cut it down by dropping anything obviously outside the range on gross margin, which is visible on the income statement in seconds.
4. Keep the candidates inside your bands on all three. Aim for at least five names. If you have fewer than three, widen the bands and note that you did.
5. Compute the median multiple of the survivors and the range. Apply the median, quote the range.

The output you keep is not a number, it is a short table: your subject's three figures, the surviving peers with theirs, and the multiple range. That table is the thing you look at again in six months when the position has moved and you are trying to remember what you thought.

When the sector set is the better one

Two situations where I go back to the sector list, because the economic approach is not universally superior.

The first is when the economics are the thing changing. If your subject is midway through a shift from a capital heavy model to a lighter one, the economic peers you select on trailing figures are the peers for the company it used to be. The sector set, for all its noise, at least contains the businesses it is becoming. In that case build both sets, apply both, and treat the gap between them as a rough measure of what the transition is worth if it completes.

The second is thin coverage. If economic screening leaves you with two names, the median of two is not a statistic and the tight range is an illusion produced by a small sample. Widen the bands, take the larger set, and be honest in your notes that the comparison is loose. A loose comparison you have labelled as loose is far more useful six months later than a precise looking figure computed off two companies you no longer remember choosing.