Taleb's barbell — safe core + convex sliver, vs the "mediocre middle".
The barbell, not the caricature: hold the vast majority in safe assets (bounded downside) and a tiny
sliver in convex bets (unbounded upside), and avoid the mediocre middle (a 60/40). Antifragility =
you can't lose much, and you win in the tails. Does it actually beat the middle on a tail basis? This
tests it — including the honest comparison to deploying the same convexity as an overlay on a real
core (the contextual_rescue
result). Alpaca SIP daily, total-return; daily-rebalanced; cash = BIL.
CONDITIONAL KEEPER — Taleb's thesis survives: convexity done right beats the middle on risk-adjusted and tail terms, at the cost of absolute return. The two-wing barbell has the best Sharpe, the best Calmar, a positive skew, and a −4% drawdown — it dominates the 60/40 on everything except raw CAGR, which is the trade you're knowingly making.
| book | Sharpe | CAGR | maxDD | worst mo | skew | Calmar |
|---|---|---|---|---|---|---|
| two-wing 85 BIL / 7.5 UPRO / 7.5 VIXM | +1.58 | +4.2% | −4% | −1.6% | +0.63 | +0.98 |
| offensive 90 BIL / 10 UPRO | +1.12 | +5.9% | −11% | −3.8% | −0.40 | +0.51 |
| defensive 90 BIL / 10 VIXM | +0.35 | +1.1% | −10% | −1.6% | +1.72 | +0.11 |
| offensive 80 BIL / 20 UPRO | +0.92 | +9.4% | −22% | −8.1% | −0.40 | +0.43 |
| — middle: 60/40 (SPY/IEF) | +0.93 | +9.7% | −21% | −7.4% | −0.24 | +0.46 |
| — SPY | +0.88 | +15.0% | −34% | −12.6% | −0.35 | +0.44 |
| — overlay: 90 SPY / 10 VIXM | +0.98 | +13.1% | −25% | −7.7% | +0.03 | +0.52 |
2016-01 … 2026-07 (10.5y), daily-rebalanced.
- The antifragile signature is real. The defensive wing (cash + long-vol) has a +1.72 skew and is green in every crisis (COVID +6.0%, rate shock +0.8%, tariffs +1.8%) — it structurally profits from disorder. The two-wing version adds a levered-equity sliver for upside and still holds a −4% max drawdown and +0.63 skew. You genuinely cannot lose much, and you win in the tails.
- It beats the middle where Taleb says it should. Two-wing barbell vs 60/40: Sharpe +1.58 vs +0.93, Calmar +0.98 vs +0.46, skew +0.63 vs −0.24 — the "mediocre middle" is dominated on every risk-adjusted and tail measure. The barbell's bet (concentrate risk in the extremes, hold nothing in the middle) pays.
- The cost is honest and large: absolute return. +4.2%/yr vs the middle's +9.7% and SPY's +15.0%. Because 85–90% sits in cash, you need to deploy far more capital — or lever the barbell — to earn meaningful dollars, and levering reintroduces the risk you removed.
Cash-heavy books flatter the Sharpe ratio (a tiny vol denominator), so read the +1.58 alongside the +4.2% CAGR — Calmar (+0.98, return per unit of drawdown) is the more honest headline and it's still the best in the field. And in a bull decade you give up a lot by sitting in cash.
The higher-return cousin is the overlay, not the barbell: 90% SPY + 10% VIXM (last row) keeps +13.1%
CAGR and flips skew to ~0 at a better Sharpe than SPY — that's contextual_rescue/bulkhead territory,
convexity bolted onto a growth core instead of a cash core. Barbell = convexity for a preservation-first
mandate; overlay = convexity for a growth mandate. Both beat the naked middle; which you pick is the
mandate.
Replace the ETF convex wings with actual OTM options (Massive/OPRA) — the true Taleb barbell is cash + far-OTM calls/puts, whose convexity (and premium cost) the VIXM/UPRO proxies only approximate — to measure the real net-of-premium antifragility across a severe crash.
export ALPACA_KEY_ID=... ALPACA_SECRET_KEY=...
python3 research/barbell_1.pyBlaque Baux — a Carter Warrens research initiative. Research only; not investment advice. The verdict falls where the data puts it — the middle is comfortable and the barbell is correct.