If you ask a retail trader what their money costs, they’ll quote their broker’s margin rate and stop thinking. Nobody shops for the loan because they don’t know the loan exists as a market. It does: a box spread on SPX is a bond you assemble from four options, repayable in cash at a date you pick, priced continuously by the deepest option market on earth. On November 12, 2026, I pulled the entire SPX chain from CBOE’s free delayed feed — 31,016 contracts, no API key — and measured what it charges. The short end printed 3.86%. The long end, 435 days out, printed 5.16%, versus roughly 5.39% for the first tier of Interactive Brokers’ margin schedule. The gap is thin, and the interesting part is exactly where it opens and closes.

This article is the full measurement: why naive box math lies on a delayed feed, the parity regression that fixes it, the tested scanner, and an honest accounting of when a box loan beats margin — and when it’s just a slower way to pay dealer rent.

A Box Spread Is a Loan, Not a Strategy

Buy a call at strike K1, sell a put at K1. That synthetic is long the index from K1. Now do the mirror image at K2 — sell the call, buy the put — and you’re short the index from K2. Net payoff at expiry: exactly K2 − K1 in cash, regardless of where the index lands. A long box costs less than its width, and that difference is interest:

cost = (K2 − K1) · e^(−r·T)

So buying a box is borrowing: you pay cost today, repay K2 − K1 at expiry. Selling a box is lending: you collect a credit today, repay the width later. The implied rate falls out of the pricing directly. Because a box has zero directional exposure, zero vol exposure, and a hard cash payoff, its rate is the purest credit instrument retail can touch — it’s why people call it “the options market’s bond.”

One mechanism detail decides everything: settlement style. American-style equity options can be exercised early, and the short leg of your box can get assigned before expiry, converting a clean loan into a margin position you didn’t want. SPX options are European-style, cash-settled — no shares, no early assignment. The loan stays a loan until the expiry date. The cash-flow sign never surprises you.

The cautionary tale is famous. In 2019, a WallStreetBets user known as 1R0NYMAN sold boxes on equity options (USO) with ~$5,000 of capital, believing they were riskless. Early assignment on the short legs turned his synthetic long into real leverage during an oil crash; the account went from a ~$58k realized loss to a hole around $700,000. Robinhood banned equity box spreads afterward and, per its own support docs, now supports index options like SPX. The lesson isn’t “boxes are dangerous.” It’s “boxes on European-settled index options are a different instrument than boxes on American-settled ETFs.” Same name, opposite failure modes.

The Measurement Problem: Delayed Quotes Will Lie to You

The plan sounds trivial: pull quotes, compute (width/cost)^(365/T) − 1 for every strike pair, take the minimum. My first version did exactly that against CBOE’s delayed quote feed — a 13.7 MB JSON of every SPX, SPXW (weekly) and SPXO contract, free, no key:

req = urllib.request.Request(
    "https://cdn.cboe.com/api/global/delayed_quotes/options/_SPX.json",
    headers={"User-Agent": "research/1.0"})
raw = json.load(urllib.request.urlopen(req, timeout=60))

The output was garbage, and the garbage is instructive. For the 2026-11-19 expiry (7 days out), the scanner reported a box “costing” $53.10 to repay $50 — a negative rate of −95.66% — and elsewhere boxes pricing at 186% APR. Physically impossible. Two things were happening:

  1. Stale-quote noise. The feed is ~15 minutes delayed and, off-hours, marks drift from where trades printed. Check any ATM strike: SPX261218C07505000 shows bid 213.60 / theo 213.51 / last trade 239.30. The last field is yesterday’s print; bid/ask are vendor-computed marks. Each strike’s mid carries ±5 points of slop.
  2. Noise amplification. A box subtracts four prices; ±5 points per leg is ~10 points on a box that differs from its width by only a few points at short DTE. At 7 days, 10 points on a $50 box is a 20% rate error — the short end, where numbers should be tightest, is where naive math explodes.

Taking the minimum rate across pairs makes it worse: minimum-of-noise systematically picks the most corrupted quotes. Deep ITM strikes were the worst offenders — wide junk marks, near-zero open interest.

The Fix: Regress Put-Call Parity

Put-call parity for European options states:

C(K) − P(K) = S − K · e^(−r·T)

The left side, the “synthetic,” is linear in strike with slope −e^(−r·T). Instead of diffing four noisy numbers per box, compute the synthetic for every liquid strike in the expiry and fit a line:

def ols(xs, ys):
    n = len(xs)
    mx, my = sum(xs)/n, sum(ys)/n
    sxx = sum((x-mx)**2 for x in xs)
    sxy = sum((x-mx)*(y-my) for x, y in zip(xs, ys))
    b = sxy/sxx                      # slope = -e^(-rT)
    a = my - b*mx
    ss_tot = sum((y-my)**2 for y in ys)
    ss_res = sum((y-(a+b*x))**2 for x, y in zip(xs, ys))
    return a, b, 1 - ss_res/ss_tot   # a, b, R^2

then r = −ln(−b)/T. Averaging across 50–180 strikes crushes the per-strike noise, and R² becomes a data-quality gate: if parity doesn’t hold to three nines in an expiry, the quotes are too corrupted to use and the expiry gets dropped instead of poisoning the output. Filters that kept the scan honest: strikes within ±15% of spot, both legs two-sided with open interest ≥ 50, R² ≥ 0.995.

The Tested Scanner

Full code, ran live against the production endpoint on 2026-11-12, Python 3.12, exit 0:

#!/usr/bin/env python3
"""
Measure the rate SPX options charge for box-spread financing.
Core idea: a long box spread repays exactly (K2-K1) at expiry, so
    cost = (K2-K1) * e^(-r*T)
Naive per-quote box math is garbage on a delayed retail feed: each strike's
mid carries +/-5 points of stale-quote noise. Fix: put-call parity says
    C_mid - P_mid = S - K * e^(-r*T)
so regress synthetics on strikes per expiry:
    synth(K) = a + b*K,   b = -e^(-r*T)   ->   r = -ln(-b) / T
R^2 gates data quality. Data: CBOE delayed quotes (free, no API key).
Tested 2026-11-12, Python 3.12, exit 0.
"""
import json, math, re, urllib.request
from collections import defaultdict
from datetime import date

URL = "https://cdn.cboe.com/api/global/delayed_quotes/options/_SPX.json"
TODAY = date(2026, 11, 12)     # frozen for reproducibility
SOFR = 0.0389                  # FRED SOFR, 2026-11-10
T13W = 0.0394                  # 13-week T-bill coupon equiv., 2026-11-10
IBKR_TIER1 = SOFR + 0.015      # benchmark + 1.5% on the small-balance tier
MIN_OI = 50                    # liquidity floor per leg
MONEYNESS = 0.15               # strike window around spot
MIN_R2 = 0.995                 # regression quality gate

PAT = re.compile(r"^(SPXW|SPX)(\d{6})([CP])(\d{8})$")

def fetch():
    req = urllib.request.Request(URL, headers={"User-Agent": "research/1.0"})
    with urllib.request.urlopen(req, timeout=60) as r:
        return json.load(r)

def parse(raw):
    """-> {(root, expiry): {strike: {'C': (bid,ask,oi), 'P': ...}}}"""
    book = defaultdict(dict)
    for o in raw["data"]["options"]:
        m = PAT.match(o["option"])
        if not m:
            continue
        root, ymd, cp, strike = m.groups()
        exp = date(2000+int(ymd[:2]), int(ymd[2:4]), int(ymd[4:6]))
        k = int(strike)/1000.0
        if o["bid"] > 0 and o["ask"] > 0:          # two-sided quotes only
            book[(root, exp)].setdefault(k, {})[cp] = (o["bid"], o["ask"],
                                                       o["open_interest"])
    return book

def ols(xs, ys):
    n = len(xs)
    mx, my = sum(xs)/n, sum(ys)/n
    sxx = sum((x-mx)**2 for x in xs)
    sxy = sum((x-mx)*(y-my) for x, y in zip(xs, ys))
    b = sxy/sxx
    a = my - b*mx
    ss_tot = sum((y-my)**2 for y in ys)
    ss_res = sum((y-(a+b*x))**2 for x, y in zip(xs, ys))
    return a, b, 1 - ss_res/ss_tot if ss_tot > 0 else 0.0

def main():
    raw = fetch()
    spot = raw["data"]["current_price"]
    book = parse(raw)
    print(f"SPX spot {spot:.2f} | benchmarks: SOFR {SOFR:.2%} | "
          f"T-bill {T13W:.2%} | IBKR tier-1 ~{IBKR_TIER1:.2%}")
    print(f"{'root':<5}{'expiry':<12}{'dte':>4} {'legs':>5} "
          f"{'implied r':>10} {'R^2':>7}")
    out = []
    for (root, exp), by_k in sorted(book.items()):
        dte = (exp - TODAY).days
        if not 14 <= dte <= 550:
            continue
        ks = [k for k in sorted(by_k)
              if spot*(1-MONEYNESS) <= k <= spot*(1+MONEYNESS)
              and "C" in by_k[k] and "P" in by_k[k]
              and min(by_k[k]["C"][2], by_k[k]["P"][2]) >= MIN_OI]
        if len(ks) < 8:
            continue
        xs, ys = [], []
        for k in ks:
            q = by_k[k]
            xs.append(k)
            ys.append((q["C"][0]+q["C"][1])/2 - (q["P"][0]+q["P"][1])/2)
        a, b, r2 = ols(xs, ys)
        if b >= 0 or r2 < MIN_R2:
            continue                     # parity broken -> unusable expiry
        r = -math.log(-b)/(dte/365.0)
        if 0 < r < 0.20:
            out.append((dte, root, exp, len(ks), r, r2))
    out.sort()
    for dte, root, exp, n, r, r2 in out:
        print(f"{root:<5}{exp.isoformat():<12}{dte:>4} {n:>5} "
              f"{r:>10.2%} {r2:>7.4f}")

if __name__ == "__main__":
    main()

The Curve, Measured

Real output from the 2026-11-12 run (spot 7,608.85; excerpts of the 33 clean expiries, R² = 1.0000 on every row that passed the gate):

root  expiry       dte  legs  implied r     R^2   vs SOFR
SPXW  2026-11-27    15    62     3.86%   1.0000    -0.03%
SPXW  2026-12-01    19    18     3.36%   1.0000    -0.53%
SPX   2026-12-18    36   178     4.71%   1.0000    +0.82%
SPXW  2026-12-31    49   156     4.88%   1.0000    +0.99%
SPX   2027-01-15    64   157     4.82%   1.0000    +0.93%
SPX   2027-04-16   155    71     4.89%   1.0000    +1.00%
SPX   2027-07-16   246    58     5.00%   1.0000    +1.11%
SPX   2027-11-19   372    30     5.09%   1.0000    +1.20%
SPX   2028-01-21   435    17     5.16%   1.0000    +1.27%

Three facts fall out of this curve, and they’re the whole business case:

1. The short end is the risk-free rate. 15-day money prices at 3.86%, against a 3.94% 13-week T-bill (Treasury, Nov 10) and SOFR at 3.89% (FRED, Nov 10). Weeks-long collateralized borrowing clears at the government curve. Paying more for short-term leverage means donating spread to somebody.

2. The long end is SOFR plus ~120bp of dealer rent. Term money out to 14 months clears at 5.0–5.16%. That premium over funding benchmarks is the compensation dealers extract for warehousing the other side. It’s sticky — it’s what the machine charges, not a mispricing.

3. The comparison that matters. IBKR’s own pages say margin loans “start at 4.38%” — that’s benchmark + 0.5%, available only at the largest balance tiers — and the schedule “begins at Benchmark plus 1.5%,” i.e. ~5.39% on small balances (benchmark = 3.89%). So at retail scale: a ~435-day box at 5.16% beats a ~5.39% blended margin rate by ~23bp, and locks the rate while margin floats with benchmark. On a $50,000 loan that’s ~$117/year — real, but small. The bigger asymmetries are structural:

When the Box Actually Wins

Blunt base case: if your broker already gives you benchmark + 0.5% and you only need floating money for weeks, margin wins — four legs each cross a 1–2 point spread, so a round trip costs ~40–60bp of notional plus commissions, eating any 20bp rate edge. The box is not a free lunch against a good broker. It wins in specific structures:

  • You need a fixed rate for a known horizon. Margin floats with benchmark. A box is a zero-coupon fixed-rate loan — lock 435-day money at 5.16% today and the number is dead until expiry. Rate certainty is something margin cannot sell at any price.
  • Your loan size falls in a bad broker tier. Tiered margin schedules punish the middle. The market doesn’t tier — the curve above holds whether you borrow $15,000 or $150,000, subject to your broker’s box requirements.
  • You’re the lender. The quieter edge. Sell a cash-secured box and you collect the curve: ~5.1% on term money, versus IBKR’s 3.38% on uninvested cash and T-bills at 3.94%. Same cash backing, defined maturity, and the counterparty risk is the OCC’s, not a fund’s.
  • Rate-hike hedging. If you hold floating-rate debt anywhere and expect the benchmark up, long boxes are a pure rates short — the curve above shows exactly what’s priced in (the market currently expects roughly flat-to-slightly-higher short rates out to 2028).

The Risks, Without Makeup

  • Execution on four legs. The measured curve is mid-market. Filling all four legs at touch, or better, via a native box/order combo at a broker that supports them (IBKR does), is the difference between the printed rate and your rate. On exotic expiries the displays thin out — the scan itself flagged short-dated junk it refused to price.
  • Broker collateral rules. Most brokers require the repayment amount held as collateral even though the payoff is capped — the “loan” doesn’t free your whole account the way margin does. It’s financing for people who already have assets, not credit for people who don’t.
  • Tax treatment is not settled. The interest component of a box loss can be characterized as capital loss rather than interest expense depending on jurisdiction and structure. Talk to an accountant who reads Section 1256 filings for breakfast before scaling this.
  • The assignment trap is venue-specific. The whole instrument is safe because SPX is European and cash-settled. Do this on American-style equity options and you’re rebuilding 1R0NYMAN’s account, leg by leg.
  • Commissions. Four contracts per unit of loan, per round trip, charged per leg — factor ~0.15–0.30% of notional all-in on typical retail schedules.

Run the scanner weekly and you have a rate monitor: the day the long box drops under your broker’s blended tier, financing went on sale — the day the short-box yield beats your cash sweep by more than friction, idle cash has a better job. The point isn’t that one instrument dominates. It’s that the price of money is quoted, in public, every day — and almost nobody reads the tape.

Building screening infrastructure like this — quote ingestion, parity-fitted curves, data-quality gates that refuse to print lies — is exactly the kind of system we engineer at Solvinc. If you want this wired to your broker’s fee schedule with alerts on your specific break-even, email [email protected].