Research Paper · SPX 0DTE Iron Fly

From backtest to live — where the edge really lives

A systematic study of 1,020 sessions reveals that the Iron Fly's entire edge sits inside 10% profit-target exits priced at theoretical mid — a price that is not fillable in live SPX markets.

Period: May 2022 – June 2026 Sessions: 1,020 backtest · 94 paper · 5 live Underlying: SPX (SPXW, PM settle) Published: June 2026
Paper bot avg P/L
+$0.66
per trade · t = 5.06
Live avg P/L (5 trades)
−$0.35
0 of 5 PT fills achieved
Round-trip slippage
$1.75
bid/ask model, 4 legs
Backtest at real fills
−$1.53
vs +$0.22 at mid
Abstract

This paper documents the full research arc of a 0DTE SPX Iron Fly strategy — from backtesting through automated paper-trading to live execution. The strategy (short ATM straddle + 10-point wings, entry 10:00 ET, 10% profit target or 13:00 ET time-stop) shows positive expectancy at mid-price fills (+$0.22/trade, 1,020 sessions). The automated paper-trading bot recorded +$0.66/trade over 94 trades at t = 5.06 — statistically significant. Five live trades produced −$0.35/trade with zero profit-target fills. Analysis reveals that $1.75/trade of bid-ask slippage on four-leg SPX combos eliminates the entire edge. At real prices the strategy loses −$1.53/trade. The paper concludes that any 0DTE multi-leg strategy must be backtested at real bid/ask prices before deployment — mid-price backtests overstate performance by $1–2 per trade.

Section 01

Thesis & Research Goals

Thesis

0DTE SPX options decay rapidly in the final trading hours before PM settlement. An iron fly centred on the at-the-money strike at market open captures maximum available premium and benefits disproportionately from intraday theta decay. A structured profit-target exit should produce positive expected value across market regimes.

Research goals

  1. Establish the strategy's historical edge via rigorous backtesting on real options data.
  2. Validate that edge in a paper-trading environment using live market data.
  3. Quantify the divergence between simulated and live performance through real trading.
  4. Identify the root cause of any divergence and determine live viability.
  5. Test alternative structures that are robust to real-world execution costs.
Spoiler

Steps 1–2 showed strong results. Steps 3–4 revealed a structural problem with the fill model, not the strategy thesis. This paper is an honest account of that discovery.

Section 02

Methodology

Data

All backtesting uses SPXW 0DTE option chains from Databento CBBO-1m (consolidated best bid/offer, 1-minute snapshots). Coverage: 9:55 AM–1:09 PM ET daily. 1,021 sessions were available from May 2022–June 2026; one was skipped for missing data. For 332 sessions, 1-second resolution leg data (CBBO-1s) was available, enabling accurate profit-target timing. Daily settlement prices come from Yahoo Finance (^GSPC close).

Strategy — Iron Fly V1

A four-leg defined-risk structure on SPXW 0DTE contracts, entered at 10:00 ET:

  • Short call at ATM strike
  • Short put at ATM strike (forming a short straddle)
  • Long call at ATM + 10 pts (upside protection)
  • Long put at ATM − 10 pts (downside protection)

ATM is determined daily by put-call parity at the entry minute:

ATM = round( median( K + C_mid − P_mid ) / 5 ) × 5
Entry
10:00 ET
Open + 30 min
Wings
±10 pts
Max risk: 10 − credit
Profit target
10%
Close at 90% of credit
Time-stop
13:00 ET
3-hour max hold
Avg credit
$8.64
Mid price at entry
Max loss
$1.36
Wing − credit

Fill Models

Two fill models are compared throughout this paper. The gap between them is the central finding.

  • Mid-price — entry and exit at the theoretical midpoint of the bid-ask spread. This is what paper-trading systems record and what backtests produce by default.
  • Real bid/ask — sell shorts at bid, buy longs at ask at entry; buy back shorts at ask, sell longs at bid at exit. Approximates realistic fills on a multi-leg SPX combo order.
Section 03

Backtesting — 1,020 Sessions

Full results (10% PT or 13:00 TS)

May 2022 – June 2026. Both fill models side by side:

Metric Mid price Real bid/ask
Sessions1,0201,020
Total P/L+$227.25−$1,555.65
Avg P/L per trade+$0.223−$1.525
Win rate61.8%15.6%
Avg win+$0.81+$0.24
Avg loss−$0.73−$1.85
Payoff ratio1.11×0.13×
PT exits358 (35.1%)
TS exits662 (64.9%)
Round-trip slippage$0$1.75 / trade
Critical finding

$1.75 of round-trip bid-ask slippage converts a +$0.22/trade mid-price edge into −$1.53/trade. Win rate collapses from 61.8% to 15.6% because virtually every small profit becomes a loss after paying the spread.

Time-stop-only baseline (no PT)

Re-running with profit-targeting disabled models the live experience where no PT fill can be achieved:

Total P/L (mid)
−$33.83
1,020 sessions
Avg P/L
−$0.033
Mid fills, TS-only
Win rate
50.7%
Essentially a coin flip
Key insight

The entire +$0.22/trade mid-price edge lives in the 35% of trades that exit at the 10% profit target. Strip those out and the strategy is flat at mid — and solidly negative at real fills. The question for live trading is not "does the strategy have edge?" but "can PT exits be filled at mid?"

Section 04

The Gap — paper vs reality

The divergence between paper and live performance is not a timing problem or a broker issue. It is structural: SPX 0DTE options carry bid-ask spreads of $0.20–$0.50 per leg. Four-leg combos accumulate $1.50–$2.00 of round-trip slippage. Any profit target smaller than this slippage is unreachable in practice.

Paper fill (mid)
+$0.66
per trade · paper bot
$1.75
slippage
Real bid/ask
−$1.53
per trade · backtest real

Performance across all layers

Every data point we have, from theory to live:

Layer
Mid fills
Real fills
Backtest (1,020 sessions)
+$0.22
−$1.53
Backtest — TS-only, no PT
−$0.03
~−$0.43
Paper bot (94 trades, t = 5.06)
+$0.66
Live trades (5 trades, Jun 2026)
−$0.35

Live sample is 5 trades; insufficient for statistical conclusions. "Real fills" for TS-only baseline is estimated (mid −$0.40). Paper bot does not track real fills — only mark price.

Where the $1.75 comes from

Opening the trade
Sell short call (at bid)−$0.15
Sell short put (at bid)−$0.15
Buy long call (at ask)−$0.08
Buy long put (at ask)−$0.08
Open subtotal −$0.46
Closing the trade
Buy back short call (at ask)−$0.30
Buy back short put (at ask)−$0.30
Sell long call (at bid)−$0.35
Sell long put (at bid)−$0.34
Close subtotal −$1.29
Total round-trip

Opening + closing = −$1.75 per trade (Databento CBBO model). Close-side slippage is higher because ATM options in the final hour carry wider spreads as delta moves around.

Section 05

Paper Trading Bot

An automated bot was deployed via the Tastytrade API to paper-trade the V1 rules in real time. It enters at 10:00 ET (14:30 UK BST), monitors live mark prices every few seconds for the 10% PT trigger, and time-stops at 13:00 ET. All fills are recorded at mark price (theoretical mid).

Closed trades
94
Dec 2025 – Jun 2026
Total P/L
+$62.37
Mark price fills
Avg P/L
+$0.66
Per trade
Win rate
74.5%
70 wins / 24 losses
t-statistic
5.06
Significant (> 2.0)
Profit factor
3.73×
Avg win / avg loss

The paper bot's results are statistically significant — t = 5.06 across 94 trades. The edge is real in simulation. The bot records a fill at mark price the moment a PT trigger is detected, even when no real order is placed or tested. This is exactly the mid-price model.

What paper confirms

The strategy has positive expected value when priced at mid. The +$0.66/trade at t = 5.06 across 94 trades is not noise. What it does not confirm is whether those PT trigger prices correspond to fillable live orders.

Section 06

Live Trading Results

From 22 June 2026, the strategy was executed with real orders at Tastytrade — same entry time, same strike selection, same management rules as the paper bot. Five trades completed before this paper was written.

Date ATM Credit Debit P/L Exit
22 Jun7530$8.45$10.00 −$1.55 Time-stop · chased close
24 Jun7390$9.00$8.65 +$0.35 Time-stop · walked exit
25 Jun7345$9.35$9.10 +$0.25 Time-stop · walked exit
26 Jun7355$9.15$9.45 −$0.30 Time-stop · walked exit
29 Jun7395$9.10$9.60 −$0.50 Time-stop · phantom PT
Total $9.01 avg$9.36 avg −$1.75 0 of 5 PT fills

29 June — the phantom PT case

On 29 June 2026, the trader and the paper bot both entered the same iron fly at 15:02 UK at $9.10 credit. Three minutes later the paper bot recorded a 10% PT exit. The live position stayed open for over three hours, closing at the 18:25 time-stop at $9.60 debit for −$0.50.

The bot's PT triggered when mark price touched $8.19 (= 90% of $9.10). But to close four legs live requires buying back two short legs at their ask prices and selling two long legs at their bid prices. The effective close cost at that moment was approximately $8.59 — $0.40 above mid. The threshold was visible on screen but not accessible in the live order book.

Root cause

A 10% PT on $9.10 credit requires closing at $8.19 debit. At 10:15 ET on 0DTE SPX, four ATM legs carry ~$0.20–$0.35 bid-ask each. The minimum realistic close is mid + ~$0.40, or ~$8.59. The PT threshold is structurally $0.40 below the fillable price. The 10% PT is systematically unreachable live.

Section 07

Alternative Structures Tested

Having established that Iron Fly V1 fails at real fills, two alternative structures were backtested against the same dataset.

0DTE Iron Condor — wing width comparison

Short strikes at ±1σ expected move (≈20 delta), long wings at varying widths, 25% profit target or PM settlement. Key finding: slippage is driven by the short strikes, which stay the same regardless of wing width. Wider wings increase credit at no additional slippage cost:

Wing widthAvg creditSlippageSlip %P/L at midP/L real
$5 wide$1.31 $0.3124% +$0.031−$0.275
$10 wide$2.34 $0.3013% +$0.076−$0.223
$25 wide$4.16 $0.317% +$0.353+$0.040
Key insight

The $25-wide condor is the first structure that survives real fills (+$0.040/trade). But its estimated t-statistic is ~0.20 across 986 sessions — not statistically proven. The payoff ratio is poor (0.29×): average loss is ~$11.92 against average win of ~$3.49, requiring roughly four wins to recover one loss.

Jade Lizard — multi-DTE

Paper-traded from May 2026: short put + $20 put wing + short call + $5 call wing, held 5/7/10 DTE to expiration, 25% profit target. With multi-day DTE, options close while still carrying meaningful time value — better liquidity than 0DTE closes.

VariantTradesTotal P/LPer tradeWin ratet-stat
JL 10-DTE15+$15.02+$1.0093.3%~0.9
JL 5-DTE18+$14.87+$0.8388.9%~0.7
JL 7-DTE15+$11.35+$0.7686.7%~0.8

t-statistics are well below 2.0 — results not yet distinguishable from a lucky streak. The payoff ratio (avg win ~$2 vs avg loss ~−$12) presents the same high-WR / fat-tail shape that caused the 20-delta iron condor to be retired. More data required before conclusions can be drawn.

Section 08

Conclusion

The SPX 0DTE Iron Fly V1 has a genuine statistical edge in simulation. The paper bot's +$0.66/trade at t = 5.06 across 94 trades cannot be explained by luck. The 1,020-session backtest confirms positive mid-price expectancy. The strategy's thesis — that 0DTE premium decays predictably — is sound.

But the edge lives entirely in 10% profit-target exits priced at theoretical mid, a price that is not fillable on four-leg SPX combos. Five consecutive live trades confirm: zero PT fills achieved, all positions forced to time-stop walk-closes. The live result (−$0.35/trade) matches the time-stop-only mid baseline (−$0.03/trade) once exit slippage is included. The research is internally consistent.

Three paths forward

Path 01
Widen the structure
Move to a $25-wide iron condor. Slippage drops to 7% of credit; real-fill P/L turns marginally positive. Edge not yet proven but structure is viable on paper.
Path 02
Raise the profit target
A 25–35% PT closes legs worth $2–3 each rather than $0.50. Liquidity is substantially better at those price levels. Lower win rate is the trade-off.
Path 03
Redesign the exit
GTC limit orders set at entry instead of chasing at time-stop. Or leg out individually rather than as a combo. Either approach reduces close-side slippage.
Final verdict

The Iron Fly V1 is statistically proven in simulation but structurally broken in live execution. The paper bot's +$0.66/trade at t = 5.06 is real — it simply exists at a price that cannot be realised in the live market.

This is not a failure of the thesis. It is a failure of the profit-target design relative to SPX 0DTE market microstructure. The strategy must be redesigned, not abandoned.

Any future 0DTE multi-leg strategy must be backtested at real bid/ask prices before deployment. Mid-price backtests on SPX structures overstate performance by $1–2/trade and should be treated as theoretical upper bounds only.

Data: Databento CBBO-1m/1s (SPXW 0DTE, 2022–2026) · Yahoo Finance ^GSPC settlement · Tastytrade mark-price fills (paper) · Manual brokerage records (live trades, Jun 2026).

Disclaimer: Research documents a specific paper-trading experiment. Not financial advice. All P/L figures are per SPX unit (×100 multiplier). Past performance does not guarantee future results.