TCE vs VLCC Stock Price — Why “Watching the Rate Tape” Fails
The “Average × Duration” thesis, proven with data
Published: June 26, 2026 · Stocks: Frontline (FRO), DHT Holdings (DHT)
⚠️ Disclaimer: Analytical research, not investment advice. Past performance does not predict future results. BDTI is used as a high-frequency VLCC dirty-rate proxy; long-cycle TCE figures are sourced approximations (flagged below).
TL;DR — The Verdict
“Watching spot TCE to trade VLCC stocks is bad” — the data agrees. A VLCC equity is not a leveraged bet on today’s TCE print. It is a claim on the average TCE level sustained over a duration. The spot tape is the noise; the trailing average is the signal.
Six independent pieces of evidence in this report:
| # | Evidence | Result |
|---|---|---|
| 1 | Stock vs rate correlation rises with the averaging window | FRO R² 0.12→0.37, DHT 0.20→0.50 (spot → 52-wk avg) |
| 2 | TCE peak vs stock peak (amplitude compression) | TCE peaks 5–10.6× baseline; stock peaks only ~1–3× |
| 3 | Real episode contrast (duration) | 2020 short spike → FRO +11%; 2022–24 sustained → FRO +307% |
| 4 | Simulation: same peak, vary duration | 2-wk spike ×1.0 vs 2-yr sustained ×1.82 |
| 5 | Simulation control: same duration, vary peak | $120k→$350k peak moves stock only ×1.66→×1.83 |
| 6 | Simulation: signal quality | Sustained-avg signal fwd return median +64% vs spot +10% |
Practical rule: trade the stock off the trailing 26–52-week average TCE and its duration, not the daily/weekly spot. A spike that does not persist does not pay.
0. WS vs TCE — what we are actually watching
- WS (Worldscale) is a freight-rate index quoted as a percentage of a published “flat” reference; it is how voyage charters are priced.
- TCE (Time Charter Equivalent, $/day) converts that voyage economics into a daily
vessel-earnings number:
TCE ≈ (voyage revenue − voyage costs) / round-trip days. - Spot TCE is extremely volatile — it routinely spikes 5–10× in geopolitical or storage events (2008, 2020, 2026 Hormuz) and round-trips within weeks. That volatility is exactly why trading the stock off the spot print is dangerous.
1. Proof #1 — The stock tracks the sustained average, not spot
Using real weekly data (BDTI as the dirty-rate proxy + FRO/DHT adjusted closes, 252 aligned weeks, 2020–2024), we regress the stock price level on the rate, widening the rate’s averaging window from spot to 52 weeks.
| Averaging window | FRO R² | DHT R² |
|---|---|---|
| Spot (1 wk) | 0.124 | 0.204 |
| 4-week avg | 0.133 | 0.213 |
| 13-week avg | 0.177 | 0.270 |
| 26-week avg | 0.257 | 0.375 |
| 52-week avg | 0.373 | 0.502 |
Read: explanatory power triples (FRO) / 2.5× (DHT) as we move from spot to the 52-week average. The longer the look-back, the better the rate explains the stock — the signature of a market pricing sustained earnings, not the latest spot.
Honest nuance: in 4-week changes, the spot rate change has a higher R² with the stock change (FRO 0.21) than the smoothed averages (~0). I.e. spot wiggles do jiggle the stock day-to-day — but the level (what the equity is worth) is set by the sustained average. Lead/lag is contemporaneous (best lag = 0 weeks), so the spot tape gives you no timing edge.
2. Proof #2 — How big is the TCE peak vs the stock peak?
This is the question directly asked. Across cycles we compare the TCE peak / baseline multiple against the stock peak / baseline multiple (stock multiples are real yfinance adjusted closes; TCE anchors are sourced approximations).
| Cycle | TCE peak × | FRO peak × | DHT peak × | Source (TCE) |
|---|---|---|---|---|
| 2008 demand super-spike | 7.7× | 3.0× | 0.9× | 2008 Baltic TD3C published peak ~$229-230k/day summer-2008 (Clarksons/Baltic, fact-checked |
| 2015 mini-cycle | 5.0× | 1.1× | 1.2× | 2015 avg ~$50-60k, intra-year peak ~$100k (web-verified) |
| 2020 COVID storage pulse | 10.6× | 1.2× | 1.2× | 2020 TD3C peak $264,072/day Mar-2020 (web-verified) |
| 2026 Hormuz spike | 8.4× | 1.9× | 1.6× | 2026 Hormuz peak ~$420-424k/day Mar-2026 (Lloyd’s List ‘VLCC index tops $420K’, fact-checked |
Answer: TCE peaks are enormous — 5× to 10.6× the pre-spike baseline. Stock peaks are tiny by comparison — typically 1–3×, and in two cases (2015, 2020) barely 1.1–1.3×. The market applies a massive amplitude compression to transient rate spikes: it refuses to capitalise a TCE number it does not believe will persist.
3. Proof #3 — Duration beats peak height (real episode contrast)
Two real episodes in the 2020–2024 window make the point without any modelling:
| Episode | Rate peak × | FRO move × | DHT move × | Character |
|---|---|---|---|---|
| 2020 COVID floating-storage spike | 1.45× | 1.11× | 1.25× | Huge spike, ~weeks → stock shrugs |
| 2022-2024 sustained up-cycle | 2.26× | 4.07× | 2.54× | Lower peak, ~18 months → stock multiplies |
In 2020 the COVID floating-storage event sent the rate +45% (and TD3C $/day to a sourced $264k peak) — yet FRO moved just +11% because it lasted only weeks. In 2022–2024 a lower peak that persisted ~18 months drove FRO +307%. Duration — not peak height — is what re-rates the equity. (The same logic explains why the 2026 Hormuz $400k all-time-high TD3C print coincided with a stock dip: pure spike, zero duration.)
4. Proofs #4–6 — Controlled simulation (isolating the mechanism)
A synthetic model removes confounders. Spot TCE = a mean-reverting baseline plus injected spikes; earnings see a charter-lagged trailing average; the stock prices normalised (52-week) earnings at a fixed multiple. (Deterministic, seed = 42; illustrative, not a forecast.)
4.1 Same peak, different duration → very different stock
| Episode | TCE peak | Duration | Stock re-rate |
|---|---|---|---|
| Spike | $200k | 2 weeks | ×1.0 |
| Plateau | $200k | 6 months | ×1.09 |
| Sustained | $200k | 2 years | ×1.82 |
Identical $200k peak; the 2-week spike does nothing (×1.0); only sustained elevation re-rates the stock (×1.82).
4.2 Control — same duration, different peak
| TCE peak (held 52 wks) | Stock re-rate |
|---|---|
| $120k | ×1.66 |
| $200k | ×1.76 |
| $350k | ×1.83 |
Tripling the peak ($120k→$350k) moves the stock only ×1.66→×1.83 (+~10%). Peak height barely matters; duration dominates.
4.3 Signal quality — why spot-chasing loses
Across 600 random rate paths we measure the forward 26-week stock return after two signal types:
| Signal | Fires (n) | Mean fwd 26w | Median fwd 26w | Win rate |
|---|---|---|---|---|
| Spot spike (TCE > 1.8× base) | 923 | +34% | +10% | 63% |
| Sustained avg (26w-avg > 1.4× base) | 766 | +68% | +64% | 80% |
The spot-spike signal fires on every spike — most of which are short and never re-rate the stock — so its median forward return is a weak +10%. The sustained-average signal only fires once a spike has persisted, delivering a +64% median and an 80% win rate.
5. Synthesis — why the spot tape misleads
- Earnings are an average, not a print. Vessels on voyage/charter realise a trailing average of the rate, not the spike. A 2-week $300k print barely moves quarterly EPS.
- The market capitalises normalised earnings. It discounts transient spikes because it knows they mean-revert; only duration converts a rate move into durable EPS.
- Operating leverage cuts both ways. When a high rate does persist (2022–24), the stock over-reacts (compression < 1) — the same leverage that makes spot-chasing a trap makes duration-confirmed entries powerful.
This dovetails with the project’s existing Modeling Stash sell framework, whose decisive rule is momentum + rate confirmation (sell only when price and the rate trend roll over) — i.e. it already trades the trend/average, never the spot spike. The 2026 Hormuz case (stock dipped while spot TD3C hit a $400k ATH) is the canonical “do not trade the spot tape” event.
6. What to watch instead — practical rules
- Primary gauge: trailing 26–52-week average TCE (or BDTI 52-wk avg), not spot.
- Entry: average crosses up and has held ≥ ~13 weeks (duration filter).
- Spike ≠ buy: a spot spike with no duration is noise; expect the stock to fade it.
- Exit: the average (not spot) rolls over and price breaks — per Modeling Stash momentum + rate-confirmation.
- Geopolitical dip: price down but average still high → hold/add, not sell.
7. Limitations & data sources
- BDTI proxy: BDTI is a blended dirty index; it understates pure-VLCC TD3C $/day amplitude (e.g. 2020 shows +45% in BDTI vs the sourced $264k TD3C spike). It nonetheless captures the relative spot-vs-sustained behaviour cleanly. Free BDTI series covers 2020–2024 only.
- Long-cycle TCE (2008/2015/2020/2026): sourced approximations (web-verified industry
figures + repo
modeling_stash.md), used for cycle shape. No fabricated data — estimates are flagged. - Stock multiples: real yfinance adjusted weekly closes (splits/dividends adjusted); pre-IPO cells are blank (DHT IPO Oct-2005).
- Simulation: illustrative mechanism only, fixed seed; not a price forecast.
- Small sample: only ~4 clean cycles in 20 years; treat magnitudes as directional.
- Sources: Baltic Exchange / Investing.com (BDTI), Yahoo Finance (prices), industry
press for TCE peaks (2008 ~$300–350k, 2020 $264k, 2015 ~$50–60k avg), repo
modeling_stash.md(2026 Hormuz ~$400k).
8. Two-Step Research Protocol (repo-mandated)
Step 1 — Concise Research Draft
Core conclusion: VLCC equity prices are driven by the average TCE sustained over a duration, not by spot-TCE peaks; therefore trading the stock off the spot rate tape is a losing strategy, and TCE peaks are several times larger in amplitude than the stock peaks they coincide with.
Supporting points (claim → evidence needed):
- Stock correlates better with trailing-average TCE than spot. → Need: rising R² of stock level vs rate as the averaging window lengthens. Have: FRO 0.12→0.37, DHT 0.20→0.50 (2020–2024 weekly).
- Duration, not peak height, drives the re-rate. → Need: same-peak/different-duration comparison + real episode contrast. Have: sim ×1.0 vs ×1.82; 2020 (+11%) vs 2022–24 (+307%).
- The market amplitude-compresses spikes. → Need: TCE-peak× vs stock-peak× per cycle. Have: TCE 5–10.6× vs stock ~1–3×.
Opposing / counter points (claim → evidence needed):
- Spot still matters short-term. → Need: spot-change vs stock-change correlation. Have (concedes): 4-wk change R² is higher for spot (0.21) than for averages — spot moves the stock intra-quarter, just not its level.
- BDTI ≠ TD3C; proxy may distort. → Need: pure-VLCC TD3C $/day weekly history to confirm. Status: unknown / paywalled — not fully verified here.
Step 2 — Strict Peer Review (review only; draft not rewritten)
1. Facts that need verification → now fact-checked (see §9)
- 2008 TD3C peak: draft said “~$300–350k/day”; fact-checked to ~$229–230k/day (published Baltic/Clarksons benchmark — $300k+ were outlier single fixtures). The 2008 row is corrected accordingly (now 7.7×, was 10×).
- 2026 Hormuz peak: draft “~$400k”; fact-checked to ~$420–424k (Lloyd’s List “VLCC index tops $420K”). 2026 row corrected (now 8.4×).
- DHT 2008 “×0.92”: DHT was a tiny fleet in 2008 and the GFC crushed Q4 — the multiple may reflect company-specific factors, not the rate mechanism. (Limitation retained.)
2. Logical leaps / concept substitution
- BDTI (blended dirty index) is silently substituted for VLCC TD3C TCE $/day; they are correlated but not identical, and the substitution understates spike amplitude.
- “R² of level vs rate” and “predicts forward returns” are different claims; the report is careful, but a reader could conflate explanatory R² with tradable alpha.
- Stock peak/baseline multiples depend heavily on the (analyst-chosen) baseline window; different baselines change the compression ratio.
3. Missing counterexamples / competing explanations
- Balance-sheet / dilution / dividend-policy changes (FRO recap, DHT payout) move the stock independently of TCE and are not isolated.
- A sustained-rate period usually coincides with rising consensus EPS and de-risking; the re-rate may be an earnings-revision effect, not “duration” per se.
- Survivorship: only FRO/DHT are studied; failed/merged tanker names are excluded.
4. Most important primary sources to add
- Baltic Exchange TD3C TCE $/day weekly history (2005–2026) — replace the BDTI proxy.
- Company quarterly realised TCE disclosures (FRO/DHT 10-Q/6-K) to tie rate → EPS.
- Clarksons/Gibson cycle chronologies for independent peak/duration dating.
5. Sentences that are at most speculation, not fact
- “TCE peaks are 5–10× while stock peaks are 1–3×” as a general law — it is 4 observations, directionally strong but not statistically established.
- The simulation re-rate multiples (×1.0 / ×1.82, etc.) are model outputs, not market measurements, and must not be read as predictions.
- “Operating leverage makes duration-confirmed entries powerful” — plausible mechanism, demonstrated only in-sample/in-model here.
9. Fact-Check & Open-Questions Resolution (added Jun 26, 2026)
This section resolves the questions raised by the §8 Step-2 peer review.
9.1 Verified / corrected facts
| Item | Draft claim | Verified value | Source | Action |
|---|---|---|---|---|
| 2008 TD3C peak | ~$300–350k | ~$229–230k/day (summer-08; $300k+ = outlier fixtures) | Clarksons/Baltic recaps | Corrected → 2008 row 7.7× |
| 2026 Hormuz peak | ~$400k | ~$420–424k/day (Mar-26) | Lloyd’s List “VLCC index tops $420K” | Corrected → 2026 row 8.4× |
| 2020 COVID peak | $264k | $264,072/day (Mar-20) confirmed | industry press | Unchanged |
| 2015 | ~$50–60k avg | avg ~$50–60k, intra-yr peak ~$100k confirmed | industry press | Unchanged |
The corrections do not change the conclusion: TCE peaks remain 5–10.6× baseline vs stock peaks ~1–3×.
9.2 Proxy concern resolved — BDTI vs TD3C
The peer review flagged that BDTI was substituted for VLCC TD3C $/day. Fact-check: BDTI is a Baltic basket (VLCC TD1/TD2/TD3C + Suezmax + Aframax routes) that includes TD3C and is strongly correlated with it, but dampened by the smaller, less-volatile routes/sizes. Implication: BDTI is a valid correlated proxy that understates pure-VLCC spike amplitude — so the true TCE-vs-stock amplitude compression is even larger than the BDTI-based numbers in §1/§3. This strengthens, not weakens, the thesis. (Status: was “unknown/paywalled” → resolved as a known, conservative bias.)
9.3 Methodology clarifications (logical-leap items)
- “R² of level” ≠ “predicts forward returns.” Kept distinct: the level-R² (§1) shows what the stock is worth; tradability is handled separately by the forward-return signal test (§4.3). We do not equate explanatory R² with alpha.
- Baseline-window sensitivity. Amplitude multiples depend on the chosen pre-spike baseline; we use a fixed window and report magnitudes as directional, not precise.
9.4 Counter-explanations acknowledged
- Earnings-revision vs duration. A sustained rate and rising consensus EPS are correlated; we do not fully separate them. Framing: duration is the mechanism, earnings revisions are the transmission — both point the same way.
- Balance-sheet / dividend effects (FRO recaps, DHT payout policy) and survivorship (only FRO/DHT studied) remain genuine limitations, not controlled for here.
9.5 Still open (honest unknowns)
- Full Baltic TD3C $/day weekly history 2005–2026 (paywalled) — would replace the BDTI proxy and measure VLCC-specific amplitude directly.
- Company realised-TCE disclosures (FRO/DHT 10-Q/6-K) to tie rate → EPS precisely.
Part of the VLCC-Analysis-2026 project.
Methods: tce_analysis.py, tce_simulation.py, generate_tce_charts.py. Not investment advice.