# When does a spin-off start having returns?

*When-issued market mechanics, the CRSP convention, and how RiskModels incorporates spin-off pricing — the IBM/Kyndryl case*

**Conrad Gann** · Blue Water Macro · Data Note · June 2026

---

Two careful data sources can disagree about a spin-off's returns — on exactly two kinds of days — while both being internally consistent. This note documents the mechanics behind that disagreement, states precisely when RiskModels begins using a spun-off company's data, and contrasts it with the CRSP (WRDS) convention. The worked example is the Kyndryl (KD) spin-off from IBM in November 2021.

The key insight is not a data-quality question. It comes from the mechanics of the **when-issued market** — a real, exchange-sanctioned market in shares that do not exist yet — and from where each methodology chooses to draw the line on using its prices.

---

## The when-issued market

In a spin-off, there is a gap between the day the distribution is fixed and the day the new shares are delivered. For IBM/Kyndryl:

- **Record date** — Oct 25, 2021: IBM holders of record become entitled to 1 KD share per 5 IBM shares.
- **Distribution date** — Nov 3, 2021: KD shares are delivered.
- **Regular-way trading** — Nov 4, 2021: KD trades as an ordinary listed security.

Starting Oct 22 — before any KD share existed — the NYSE opened a **when-issued market** ("KD WI"): binding trades in the *entitlement* to shares, with settlement deferred until after the distribution. Had the spin-off been cancelled, all when-issued trades would have been voided. In parallel, regular-way IBM traded with **due bills** attached through Nov 3, so IBM buyers in that window still received the KD distribution — which is why IBM's price did not drop until Nov 4.

When-issued prices are real price discovery. Volume is thinner and the participants skew institutional, but the trades are binding and arbitrage-linked to the regular-way market that follows: the last when-issued close and the first regular-way price are mechanically the same instrument.

## Where each methodology draws the line

**RiskModels** (built on exchange-tape vendor data) carries the when-issued tape under the regular ticker. KD's price history therefore begins **Oct 22, 2021** — the first when-issued print — with the first *return* on Oct 25 (the first day with a prior close). The transition from when-issued to regular-way is continuous; no special handling is applied, and none of the pre-listing history is synthesized.

**CRSP** books when-issued trading separately (its own identifier, flagged as a when-issued share class) and starts the regular security at **regular-way trading with the first return missing**. KD's first CRSP return on the main identifier is Nov 5, 2021.

Both choices are defensible. RiskModels' convention captures ten extra trading days of genuine price discovery on the new entity; CRSP's convention guarantees every return is computed within a single trading regime. The consequence is that a cross-check between the two will show "mismatches" on every day from the first when-issued print through the first regular-way day — dates on which, in CRSP's framing, the security does not yet exist.

### Kyndryl, day by day

| Date | KD close | RiskModels return | CRSP-convention return | Regime |
|---|---|---|---|---|
| 2021-10-22 | 40.75 | NaN (first observation) | — not covered | when-issued |
| 2021-10-25 | 37.85 | −7.12% | — not covered | when-issued |
| 2021-10-26 | 35.50 | −6.21% | — not covered | when-issued |
| 2021-10-27 | 33.00 | −7.04% | — not covered | when-issued |
| 2021-10-28 | 32.00 | −3.03% | — not covered | when-issued |
| 2021-10-29 | 31.50 | −1.56% | — not covered | when-issued |
| 2021-11-01 | 33.50 | +6.35% | — not covered | when-issued |
| 2021-11-02 | 31.25 | −6.72% | — not covered | when-issued |
| 2021-11-03 | 28.50 | −8.80% | — not covered | distribution date |
| **2021-11-04** | **26.38** | **−7.44%** | **missing (first day)** | **regular-way begins** |
| 2021-11-05 | 24.25 | −8.07% | −8.07% | identical from here on |
| 2021-11-08 | 23.75 | −2.06% | −2.06% | identical |

Two things are worth noting. First, the when-issued slide from 40.75 to 28.50 was genuine: the market repricing standalone Kyndryl ahead of delivery. Second, the −7.44% on Nov 4 — the day CRSP reports as missing — is a real move whose denominator happens to be a when-issued close. Discarding it would delete information, not noise.

### Not a ratio imputation

A natural suspicion is that pre-Nov-4 KD prices are synthesized from IBM via the 0.2 distribution ratio. The arithmetic rules this out:

| Date | IBM close | IBM × 0.2 (imputed would be) | Actual KD close |
|---|---|---|---|
| 2021-11-01 | 126.28 | 25.26 | 33.50 |
| 2021-11-02 | 126.18 | 25.24 | 31.25 |
| 2021-11-03 | 127.13 | 25.43 | 28.50 |

The when-issued prints sit 12–30% above any ratio-scaled IBM price, the gap changes daily, and the two series move independently — on Nov 3, KD fell 8.8% while IBM rose 0.75%. A ratio-imputed series would be perfectly correlated with the parent by construction. KD's series also starts abruptly at Oct 22 (the when-issued open), not at some arbitrary lookback depth, which is the signature of an independent tape rather than a back-filled transform.

---

## The parent's ex-date return: one day, one wedge

The second — and subtler — divergence is in the **parent's** return on the ex-date, and it persists even after the child's coverage question is settled.

IBM's raw price drop on Nov 4 was −4.94% (127.13 → 120.85), most of which is the value of the distributed Kyndryl shares leaving the stock. Any total-return methodology adds that value back. The conventions differ only in *which KD price* values the distribution:

- **CRSP** values the distribution at the child's **same-day (ex-date) close**: (120.85 + 0.2 × 26.38) / 127.13 − 1 = **−0.79%**. This measures the realized wealth change of a holder who received the shares.
- **Vendor adjustment factors** (the standard commercial-data approach, used by the tape behind RiskModels) are fixed **before the ex-date open**, so the distribution is valued at the child's price known beforehand — the when-issued close / opening indication, here ~27.8: (120.85 + 0.2 × 27.80) / 127.13 − 1 = **−0.57%**.

| Date | IBM close | KD close | RiskModels return | CRSP-convention return |
|---|---|---|---|---|
| 2021-11-01 | 126.28 | 33.50 | +0.94% | +0.94% |
| 2021-11-02 | 126.18 | 31.25 | −0.08% | −0.08% |
| 2021-11-03 | 127.13 | 28.50 | +0.75% | +0.75% |
| **2021-11-04** | **120.85** | **26.38** | **−0.57%** | **−0.79%** |
| 2021-11-05 | 123.61 | 24.25 | +2.28% | +2.28% |
| 2021-11-08 | 124.54 | 23.75 | +0.75% | +0.75% |

The entire gap is the child's own first-day move scaled by the distribution ratio:

> 0.2 × (27.80 − 26.38) / 127.13 ≈ **22 bp**

Every other day matches exactly. The wedge appears only on spin-off ex-dates, is bounded by `distribution ratio × child's first-day move ÷ parent price`, and **nets out over any multi-day window** — both series converge on the same prices immediately afterward. Neither number is an error: one is an ex-ante adjustment factor, the other an ex-post realized total return, and they answer slightly different questions on one day.

---

## Summary of conventions

| Question | RiskModels | CRSP |
|---|---|---|
| When does the child's price history start? | First when-issued print (e.g., KD: Oct 22, 2021) | First regular-way day (KD: Nov 4, 2021) |
| Child's first return | First when-issued day with a prior close | Missing on day one; first return on day two of regular-way |
| When-issued period | Included, under the regular ticker | Separate when-issued identifier |
| Parent's ex-date distribution valuation | Adjustment factor fixed pre-open (prior / when-issued price) | Child's same-day close |
| Are pre-regular-way prices imputed? | No — independent when-issued tape | n/a |

For model estimation the distinction is further insulated: a new listing enters the RiskModels universe only at a month-end rebuild, and beta estimation requires a minimum data density over a 126-day window, so the earliest (thinnest) when-issued days do not drive risk estimates. The convention difference is visible primarily on the per-ticker daily returns surface — and, as documented here, it is a difference in where two methodologies draw the same line, not a disagreement about what traded.