On a book of forty to eighty names you almost certainly carry two fair values for most positions: one built from a peer multiple and one from a discounted cash flow. They disagree. The standard desk treatment is to take the lower of the two, call that conservatism, and move on to the next name. That is a defensible position on any single stock and a bad one across a book, because the disagreement is not independent noise draw by draw. It has a sign that repeats, a size that clusters, and a loading on characteristics that already appear on your exposure report.
The check I would want run before the next quarterly review is simple to specify and takes an afternoon: compute the spread for every holding, look at the distribution rather than the individual numbers, and find out how much of it is explained by things you are already paid to have a view on.
The spread is a number, so treat it as one
Define the spread per name as the natural log of the intrinsic value divided by the multiple-based value. Logs, not percentages, because you want plus thirty and minus thirty to be symmetric and because you are going to average them. A holding where the DCF says 90 and the comps say 70 carries a spread of about plus 0.25. A holding where the DCF says 60 against comps of 80 carries about minus 0.29.
Two aggregates matter. The book-weighted mean spread tells you whether your process systematically prefers one method, which is a statement about house assumptions and not about the market. The cross-sectional dispersion tells you how reliable either number is as an input to sizing. A book with a mean spread near zero and a dispersion of 0.45 is not a book where the methods agree. It is a book where they disagree violently in both directions and the errors happen to cancel in the average, which is the worst case to discover during a drawdown.
The mechanism behind the spread is worth stating plainly because it determines what you can do about it. A multiple prices the current earnings base against a cross section that is itself repricing. A discounted cash flow prices a path and a terminal assumption. So the spread widens when the earnings base is off mid-cycle, when the terminal value carries most of the present value, and when the peer group has moved for reasons unrelated to the subject company. Those three causes call for three different responses, and averaging the two values together answers none of them.
Agreement is the rare case, not the base case
It helps to see the shape of the thing before arguing about individual names. The Company Valuation Engine runs a composite verdict across a large listed universe, and its header strip splits that universe into three buckets.
Do not borrow that verdict. The engine fuses fundamental ratios, technical momentum and sentiment into a single composite, which is not the same object as your two-method reconciliation, and the counts are a reading from one moment rather than a permanent property. What transfers is the shape. When a wide universe is scored against any fair value construct, the narrow band called agreement is thinly populated by construction, and the interesting mass sits in the tails. Run the same histogram on your own holdings and the question changes from why does this one name disagree to what characteristic do the tails share.
Regress the spread on exposures you already report
Once you have a spread per name, the test is whether it is idiosyncratic modelling error or a factor exposure with a different label. Regress the spread on four regressors you already hold in a database: the share of the discounted cash flow value sitting in terminal value, gross margin, the reinvestment rate, and net debt to EBITDA. Add sector dummies only if you have the names to support them.
Read the result honestly. A low explained share means your DCF assumptions are drifting name by name and the fix is process discipline on the model, not a portfolio-level story. A high explained share means the reconciliation gap is a duration bet, or a capital intensity bet, or a leverage bet, that has entered the book through the valuation layer rather than through a deliberate allocation decision. That is the sentence worth writing down: the disagreement is not an accident of arithmetic, it is an exposure nobody voted for.
With forty holdings the degrees of freedom are thin and the coefficients will be unstable. Two mitigations, both cheap. Pool across quarters so each name contributes several observations, and cut the specification to two regressors when the standard errors are wide. A stable two-variable result you can defend beats a five-variable result that flips sign every time you rebalance.
The half page the risk committee actually needs
Committees do not read reconciliation workbooks. They read one table and ask one question, which is whether the number changed and why. Six rows are enough.
| Line | What it answers |
|---|---|
| Book-weighted mean spread | Does house process systematically prefer one method |
| Cross-sectional dispersion | How much confidence either fair value deserves in sizing |
| Share of book with spread wider than 0.35 | How much capital sits on unreconciled ground |
| Top five contributors, weight times spread | Which positions drive the aggregate |
| Explained share from the regression | Whether this is model noise or an exposure |
| Change since last quarter, decomposed | Whether the move came from prices, estimates or method |
Add one line of falsification underneath. Something like: if next quarter the explained share falls below a stated level while dispersion holds, we treat the gap as model noise and drop the exposure framing. A metric with no stated way to be wrong will still be in the pack three years from now, being read by nobody.
When reconciliation stops and sizing takes over
Some spreads do not close and pretending otherwise wastes analyst weeks. Three cases recur. Companies whose peer set is genuinely thin, where the multiple carries a comparability error larger than the spread. Companies where maintenance capital spending is not disclosed at any useful granularity, so the intrinsic value rests on an assumption rather than a disclosure. And holding structures with large minority interests or equity-accounted stakes, where the two methods are quietly valuing different economic entities.
For those, the honest procedure is a sizing rule instead of another model iteration. Cap the position at a stated weight while the spread stays wide and unexplained, record which of the three cases applies, and revisit on a fixed date rather than when the price moves. The rule is not that a wide spread makes a position wrong. It is that a wide spread means you have less information than your position size implies, and that is a statement about capital allocation you can defend in a review when the name goes against you.
The market cap and P/E columns on any screen, including the one above, are the shallow end of this work. They tell you what the crowd pays today. The reconciliation tells you where your own two answers pull apart, and that is the part of the process a client can actually hold you to.