Sum-of-the-parts work has a structural problem that has nothing to do with the arithmetic. The arithmetic is easy and it always produces a discount, because segment multiples drawn from focused pure plays applied to segments inside a conglomerate will nearly always sum above the market capitalisation. What the arithmetic never produces is a date, and without a date a thirty percent discount is not an opportunity, it is a permanent feature of the security that you may be paid to hold or may simply own forever.
Which is why the useful question is not how large the gap is. It is what would have to happen for the gap to close, whether that thing has a mechanism behind it, and how long the mechanism takes.
Sorting gaps by the mechanism that resolves them
Ranked by how much of the outcome is inside somebody's control rather than in the market's mood, gap types fall into a fairly clear order.
At the top, a separation already in progress with a filed registration document, a stated timetable and disclosed segment financials. The mechanism exists, the counterparty is committed, and most of the remaining risk is execution and market window. The discount here is usually already narrow, and the work is about the residual rather than the concept.
Below that, a holding company trading below the market value of listed stakes it controls. This one is measurable to the cent daily, which is its appeal and also its trap. The tax on unwinding a low basis stake is real, and the after-tax net asset value is the honest denominator. A large part of what looks like a discount at many holdcos is simply the deferred tax the structure carries, and no announcement removes it.
Then a segment gap where one division inside a reporting group would command a much higher multiple standalone. Closure requires either separate reporting that lets the market see it, or an actual separation. Nothing forces either. This is the category where the most analytical effort is spent and the least closure occurs.
Then a non-core asset gap, meaning owned real estate, a minority stake, or a business line the group has publicly called non-core. There is a mechanism, which is a sale, and the useful signal is whether management has ever completed one.
At the bottom, a conglomerate discount with no announced action, no activist on the register and a controlling shareholder who benefits from the structure. There is no mechanism here at all. That discount can persist for a decade, and quite often the correct reading is that it is not a discount but the market's price for the governance arrangement.
The band on the board, and what migration into it would mean
That last point is the whole reason this article exists in the form it does. I am not going to hand you a table of closure rates by gap type, because any percentage I put in it would be a number I could not source, and a sourceless number in a valuation memo is worse than no number. What the desk can do is build the sample itself, and the build is genuinely tractable at a size of two or three hundred observations.
The schema for a closure sample that survives review
One row per gap observation, and the row has to be opened at the moment of observation rather than reconstructed later, which is the single discipline that keeps the sample honest.
Record the date, the identifier, the gap type from the ranking above, the measured discount to your own sum-of-the-parts at that date, the method used to build it including the specific comparables and multiples, whether a mechanism was present and named, and the expected time to closure you underwrote. Then leave it alone and mark it periodically.
Closure needs a definition fixed in advance, and the least gameable one is the discount narrowing to below half its opening level and staying there for two consecutive quarters. Half rather than zero, because gaps rarely close fully, and persistence, because a single quarter of narrowing is often just a rotation.
Three things will corrupt this sample if they are not handled at construction. Observations opened only on names you already owned, which selects for cases you found convincing rather than for the population. Retrospective adjustment of the original sum-of-the-parts once the outcome is known, which is undetectable in a spreadsheet and fatal to the base rate. And treating a takeover of the whole company as a closure, which is a different event with a different distribution and belongs in its own column.
Two or three years of disciplined recording gives you something no external source will, which is a closure rate specific to your own methods, your own market and your own definition of a gap. That is the number that belongs in a memo, and it is defensible precisely because you can show how it was built.
The arithmetic of waiting
Once a timeline is attached, the gap converts into a return and the return has to compete on the same terms as everything else in the book.
A name trading at seventy percent of your sum-of-the-parts offers roughly forty three percent if it converges fully. Over three years that is about twelve and a half percent a year. Over five it is about seven and a half. Over eight it is under five, which is below what the desk expects from ordinary positions carrying far less structural risk, and at that point the correct conclusion is that a thirty percent discount is not worth owning for the discount alone.
Then subtract what erodes it during the wait. Ongoing central costs at a holding structure, which are a real annual leakage from net asset value. Dilution where the parent issues stock to fund the parts. And drift in the comparables you built the parts from, which cuts both ways and is the term analysts consistently forget, since a gap measured against peers on twelve times that derate to nine has closed by arithmetic without anything at all having happened.
Underwrite the position at your expected timeline plus the historical dispersion around it from your own sample, not at the timeline management indicated. Management's timetable for a separation is an aspiration, and the distribution of outcomes around it is wide in one direction only.
The exit rule that belongs in the memo before the position exists
Every sum-of-the-parts position needs a written condition for closing it on the passage of time rather than on price, because the failure mode of this strategy is not loss, it is indefinite dead money in a name that always looks cheap.
Two clauses cover most of it. If the mechanism you underwrote is withdrawn, meaning the separation is shelved or the activist exits the register, the thesis has ended even if the discount has widened, and widening is not a reason to add. And if the underwritten timeline passes with no measurable progress against milestones stated at entry, the position converts to an ordinary holding to be judged on the operating business alone, at whatever weight that business justifies, which is nearly always smaller.
Both clauses are trivial to write at entry and close to impossible to apply later without them, because by then the analyst has lived with the name for three years and the discount has become an article of faith rather than a measurement.