Phase Space AI

Financial Model Notes

Repligen Corporation [RGEN]

Repligen Corporation [RGEN] — Financial Model Notes

As of 2026-07-29 · framework v1.5.1

Every number in this memo, its source, and how it was derived. No workbook is shipped: the valuation is a reverse DCF plus an own-multiple percentile, both of which are single Python scripts reproduced below rather than a spreadsheet. Scripts live in work/ in the build worktree and are listed in the manifest.

---

1. Source documents

Alpha Vantage: no data. Returned the 25/day rate-limit response on 2026-07-29. No consensus figure is used anywhere in this memo.

---

2. Derived line items — every step shown

2.1 TTM revenue to 2026-06-30

FY2025 total revenue          738,256   [A, C]
+ H1 2026 total revenue       398,383   [C]
− H1 2025 total revenue       351,538   [C]
= TTM revenue                 785,101

The screen's $763,339k is the same construction one quarter earlier (738,256 + 194,255 − 169,172) and is arithmetically correct — but 119 days stale at scan date and superseded on 2026-07-28.

2.2 Quarterly revenue series, with Q4 derived

US registrants do not tag Q4 separately, so Q4 = FY − (Q1+Q2+Q3). Derived Q4s:

Q4 2021 = 670,534 − (142,837+162,960+178,216) = 186,521
Q4 2022 = 801,536 − (206,400+207,633+200,741) = 186,762
Q4 2023 = 632,362 − (165,380+159,169+141,192) = 166,621
Q4 2024 = 634,439 − (153,182+158,839+154,871) = 167,547
Q4 2025 = 738,256 − (169,172+182,366+188,805) = 197,913

Cross-check, Q4-2025: derived $197,913k against the release's "fourth quarter revenue of $198 million" — ✅ ties.

Cross-check, Q4-2023: it does NOT tie, and the reason is a restatement. The Q4-2023 release (2024-02-21) reported Q4 total revenue of $155,743k and FY2023 of $638,764k. Current XBRL gives FY2023 $632,362k and a derived Q4 of $166,621k. The gap was traced to filing vintage:

FY2023 revenue, 10-K   filed 2024-02-22  =  638,764   ← as originally reported
FY2023 revenue, 10-K/A filed 2024-11-18  =  632,362   ← RESTATED, −6,402
FY2023 revenue, 10-K   filed 2025-03-14  =  632,362   ← carried forward
FY2023 revenue, 10-K   filed 2026-02-26  =  632,362   ← current

That −$6,402k reconciles to the −$6.6m FY2023 aggregate overstatement disclosed in the Item 4.02 non-reliance 8-K of 2024-09-18 (rounding and the royalty/other line account for the residual). Q4-2023 rose $10.9k thousand-scale against the disclosed +$10.7m. Every revenue figure used in this memo is the post-restatement series. See RGEN_Research.md §3.0. Anyone rebuilding this from press releases rather than current XBRL would inherit the pre-restatement numbers and overstate the 2023 base.

2.3 TTM GAAP EBIT to 2026-06-30

FY2025 income from operations   55,167   [C]
+ H1 2026                       29,916   [C]
− H1 2025                       20,474   [C]
= TTM GAAP EBIT                 64,609
TTM GAAP operating margin = 64,609 / 785,101 = 8.23%

2.4 Net cash — the figure the screen got wrong

Cash and cash equivalents           606,783   [C, 2026-06-30]
+ Marketable securities             203,666   [C]
= Cash + securities                 810,449
− Convertible notes, FACE           600,000   [A, note 13]
= NET CASH (face basis)             210,449   ← used throughout

Alternative on carrying value:
− Convertible notes, net carrying    551,046   [C]
= Net cash (carrying)               259,403

Face value is used because the obligation is $600.0m of cash at 2028-12-15 (conversion price $203.06, stock 28.6% below), not the $551.0m discounted carrying amount. This is the conservative and economically correct choice, and it is 2.4% of EV.

Screen's $36,065k = 606,783 (at 3/31: 582,650) + 0 − 546,585. Reproduced exactly; see RGEN_Research.md §1.1.

2.5 Enterprise value

EV = 145.00 × 56,427,067 − 210,449,000 = 8,181,924,715 − 210,449,000 = 7,971,475,715
   = $7,971.5m

2.6 Multiples

EV / TTM Sales          = 7,971.5 / 785.1  = 10.15x
EV / TTM GAAP EBIT      = 7,971.5 / 64.6   = 123.4x
EV / FY26E adj EBIT     = 7,971.5 / 131.0  =  60.9x   (guide midpoint $128–134m)
P / FY26E adj EPS       = 145.00 / 2.06    =  70.4x   (guide $2.03–2.09)

2.7 ROIC

NOPAT      = 64,609 × (1 − 0.22)                     =   50,395
Invested   = equity 2,112,032 + debt(face) 600,000
             − cash & securities 810,449             = 1,901,583
ROIC (GAAP)                                          =     2.65%
ROIC on FY26E adjusted EBIT 131,000 × 0.78 = 102,180 =     5.37%
Goodwill 1,104,183 + intangibles 357,622 = 1,461,805 = 76.9% of invested capital

2.8 Demonstrated revenue CAGR — all windows, so the choice is visible

3.0y  TTM Jun-23 712,052 → TTM Jun-26 785,101   = (1.10259)^(1/3)−1  =  +3.30%   ← primary
3.5y  FY2022 801,536     → TTM Jun-26 785,101   = (0.97950)^(1/3.5)−1 = −0.59%
5.5y  FY2020 366,260     → TTM Jun-26 785,101   = (2.14356)^(1/5.5)−1 = +15.00%
9.5y  FY2016 104,541     → TTM Jun-26 785,101   = (7.50999)^(1/9.5)−1 = +22.90%

TTM Jun-23 = Q3'22 200,741 + Q4'22 186,762 + Q1'23 165,380 + Q2'23 159,169 = 712,052

The 3.0-year figure is primary because it is the like-for-like construct with what the screen claimed to be computing (n = min(3, len(revs)−1)). The 5.5y and 9.5y windows both span the COVID boom and are reported for completeness, not used.

The screen's 30.5% reproduced: (141,236 / 63,443)^(1/3) − 1 = 30.57% → 30.5%, i.e. FY2014 → FY2017.

2.9 Screen's 39.1% operating margin reproduced

55,167 (FY2025 OperatingIncomeLoss) / 141,236 (FY2017 Revenues) = 39.06%  →  39.1%
15,974 (FY2016 OperatingIncomeLoss) / 104,541 (FY2016 Revenues) = 15.28%
39.06 − 15.28 = 23.78pp  →  the screen's op_margin_delta_pp of 23.8

---

3. The own-multiple series — construction

work/own_multiple.py. For each of 1,509 daily bars from 2020-07-27:

1. TTM revenue as known at that date — the sum of the four quarters whose results had been published by then, using the actual earnings-release dates (not period-end dates). This avoids look-ahead.
2. Shares (56,427,067) and net cash ($210.4m) held at today's verified values, so the series isolates multiple movement — which is what a percentile is for.
3. EV/Sales = (close × shares − net cash) / TTM revenue.

Usable window begins 2022-02-23 (the first date on which four consecutive quarters had been published from the series), giving n = 1,111. Percentile = share of days at or below today's 10.15x.

Deliberate limitation, stated: holding net cash constant back-dates today's balance sheet onto 2022 prices. Net cash was larger in 2022 (pre-2023 convert issuance the company had no debt), so the 2022–23 multiples in this series are marginally overstated relative to a true point-in-time build. That biases the current percentile upward — i.e. the true percentile is, if anything, lower than 14.5. The direction of the bias works against the conclusion drawn from it, which is why it is acceptable.

---

4. The reverse DCF

assets/reverse_dcf.py via work/implied_path.py. project_ev is deliberately terminal-only:

revenue_T    = revenue_0 × (1 + cagr)^years
terminal_EV  = revenue_T × terminal_margin × exit_multiple
PV           = terminal_EV / (1 + wacc)^years

Bisection on the single unknown (monotone in every parameter). Terminal value is 100% of EV by construction, which is why valuation.md makes the reverse DCF mandatory as the primary long-horizon output here and no forward DCF is run.

Fixed: EV $7,971.5m · revenue₀ $785.1m · 5 years · WACC 10% · terminal margin 18.6% · exit 26.2x EV/EBIT. Solved: required revenue CAGR 27.40%.

Peer-set construction for the two anchors

work/peers.py, over the 4,018 status == OK records in reports/scan_final/. Growth band ±50% around 12.5% (RGEN's corrected organic rate), i.e. 6.25%–18.75%; ev_ebit restricted to 0 < x < 200 to exclude negative-EBIT artefacts; SIC2 in {28, 38}; market cap > $1bn. n = 51. Median EV/EBIT 26.2x, median operating margin 18.6%.

Caveat carried forward, and it matters. That peer set is drawn from the same scanner whose op_margin_pct this memo has just shown to be corruptible by the annual_series tag-recency defect. Any peer that migrated its revenue tag has an inflated margin and hence a deflated ev_ebit. The direction of that bias makes the peer median EV/EBIT too LOW, so a corrected peer set would give a higher exit multiple and a lower required CAGR. The 26.2x anchor is therefore conservative against the long case in the wrong direction, and the §3.4 sensitivity extends to 40.0x — where the required CAGR is still 17.06% against 12–13% delivered. The FAIL survives the caveat with room. This is flagged as a systematic defect requiring a universe rerun, not repaired here.

---

5. The mention-frequency series

work/mention_freq.py. Regex counts over the 15 consecutive quarterly 8-K exhibit 99.1 earnings releases, Q4-2022 → Q2-2026, each stripped of HTML to plain text. Case-insensitive except OPUS. Release lengths 20,915–30,933 characters, so raw counts are comparable without normalisation; character counts are recorded in the output for anyone who wants to normalise.

Not transcripts. Earnings-call transcripts were not available in this run. Press releases are a filing-grade primary source and carry the headline bullets and the CEO quote, which is where a retired disclosure shows up first — but Q&A colour is absent and that limitation is stated rather than papered over.

---

6. Figures deliberately NOT modelled

---

7. Reproduction

work/fetch_pr.py       # 15 quarterly 8-K exhibit 99.1 releases -> /tmp/rgen_pr/
work/mention_freq.py   # the mention-frequency table
work/prices.py RGEN    # 1,509 daily bars -> /tmp/RGEN_bars.json
work/own_multiple.py   # EV/Sales history + percentile
work/peers.py          # growth-matched exit multiple + terminal margin
work/implied_path.py   # reverse DCF, both sensitivities
work/target12m.py      # 12-month target, standalone and pro forma
work/mkt.py            # momentum, realised vol, liquidity, options chain

All read only primary data (data.sec.gov, Alpaca) plus reports/scan_final/ for the peer set. work/env.sh sources credentials; no credential value is printed by any script.