Nasdaq Open Sigmoid Vertical Research

Real fixed-rule profit0.8%

Real fixed-rule PnL$844

Real fixed-rule Sharpe0.99

Split-test real profit0.9%

Verdict

The market thesis is partially supported: the first 30-120 minutes contain useful information about the same-day close and level breaks. The current best expression is a short 0DTE credit spread, not the original long debit spread.

The original direct implementation, buying cheap normalized verticals and holding to close, loses money in the modeled test. Main-rule PnL is $-1,949 with max drawdown -2.8%.

Real Alpaca option trade bars shrink the edge sharply. The fixed 60m credit-spread sample has real profit 0.8% and annualized daily Sharpe 0.99; the split-test sample is positive but only has 10 trades, so its Sharpe is not reliable.

Why Accuracy Is High

不是神奇预测，而是早盘状态已经包含很多当天信息。 这里的 raw direction test 只问一个简单问题：开盘后 30/60/120 分钟的方向，能不能判断当天收盘相对开盘是涨还是跌。QQQ 这种高流动性指数 ETF 在开盘后会快速消化隔夜信息、宏观新闻、期权 gamma/hedging flow 和机构订单，因此早盘方向经常会延续到收盘。

速度快的原因也很直接。 模型不是深度学习，也不需要逐笔 order book。每天只生成一行特征：early return、early range、VWAP gap、gap、ATR、以及相对前几天支撑/阻力位的距离。walk-forward 部分用 logistic regression，样本只有 2025 年 239 个交易日，所以几秒内就能完成。

但这个准确率不能直接等价成赚钱。 120 分钟准确率更高，是因为已经观察了更多当天走势，剩余预测距离更短；而 vertical spread 的价格会把一部分概率优势提前计入 premium。当前主规则的 modeled vertical PnL 仍然为负，说明需要真实期权 bid/ask 和严格 out-of-sample 验证。

Early Direction Signal

30 minearly direction vs close direction

70.7%

60 minearly direction vs close direction

75.3%

120 minearly direction vs close direction

77.8%

Support / Resistance Model

30 min calllevel-break AUC

74.7%

30 min putlevel-break AUC

66.5%

60 min calllevel-break AUC

77.3%

60 min putlevel-break AUC

67.6%

120 min calllevel-break AUC

80.5%

120 min putlevel-break AUC

76.8%

Split Test P&L

Fixed Rule P&L

Profit % And Sharpe Definition

Profit % is portfolio P&L divided by the $100,000 starting equity. It is not annualized. Sharpe is annualized from business-day returns, with zero returns on non-trade days between the first and last trade. For the 10-trade split test this statistic is fragile; the 56-trade fixed-rule sample is more informative.

Sample	P&L Source	Trades	Portfolio PnL	Profit %	Max DD	Sharpe	Win Rate
split_test	Modeled	10	$2,351	2.4%	0.0%	7.72	100.0%
split_test	Real intrinsic	10	$944	0.9%	0.0%	6.98	100.0%
split_test	Conservative	10	$847	0.8%	0.0%	6.97	100.0%
fixed_60m_w5_p045_c010_030	Modeled	56	$6,711	6.7%	-0.6%	5.11	91.1%
fixed_60m_w5_p045_c010_030	Real intrinsic	56	$844	0.8%	-0.7%	0.99	87.5%
fixed_60m_w5_p045_c010_030	Conservative	56	$286	0.3%	-1.0%	0.31	87.5%

Data Scope

QQQ2025Stock 1-minute barsAlpaca option 1-minute trade bars0DTE credit spreads

The primary stock cache lacks 9:00-9:29 premarket bars, so the 30-minute test is 9:30-10:00 New York time. The real option validation uses historical option trade bars, not historical bid/ask quotes.

Prediction Evaluation

Minutes	Model	Obs	Base Rate	AUC	Accuracy	Brier
30	call_level_break	139	27.3%	0.747	64.0%	0.217
30	put_level_break	163	19.0%	0.665	52.8%	0.255
30	raw_early_return_direction	239	54.0%	n/a	70.7%	n/a
60	call_level_break	141	27.0%	0.773	67.4%	0.206
60	put_level_break	159	12.6%	0.676	68.6%	0.213
60	raw_early_return_direction	239	54.0%	n/a	75.3%	n/a
120	call_level_break	139	24.5%	0.805	65.5%	0.197
120	put_level_break	160	11.2%	0.768	61.3%	0.216
120	raw_early_return_direction	239	54.0%	n/a	77.8%	n/a

Direct Vertical Rules

Minutes	Side	Width	Trades	Win Rate	Total PnL/Lot	Avg PnL/Lot	Debit Norm
30	call	1.000	20	20.0%	$-195	$-10	29.5%
120	put	1.000	36	22.2%	$-228	$-6	27.6%
120	call	1.000	19	21.1%	$-234	$-12	30.9%
30	put	1.000	35	20.0%	$-325	$-9	28.0%
60	call	1.000	15	6.7%	$-344	$-23	30.7%
60	put	1.000	15	6.7%	$-346	$-23	28.4%
120	put	2.000	44	22.7%	$-562	$-13	26.3%
60	call	2.000	28	25.0%	$-613	$-22	29.8%
30	put	2.000	50	22.0%	$-707	$-14	26.9%
30	call	2.000	36	25.0%	$-754	$-21	29.4%
120	call	2.000	33	21.2%	$-763	$-23	29.3%
60	put	2.000	25	4.0%	$-1,392	$-56	29.2%
120	put	5.000	40	22.5%	$-2,337	$-58	22.5%
60	put	5.000	35	14.3%	$-2,427	$-69	24.1%
30	call	5.000	52	23.1%	$-2,751	$-53	25.0%
60	call	5.000	50	22.0%	$-2,875	$-57	25.3%
120	call	5.000	46	19.6%	$-2,882	$-63	24.6%
30	put	5.000	59	23.7%	$-3,098	$-53	24.4%

Exploratory Rule Sweep

Minutes	Side	Width	Min Prob	Debit Min	Debit Max	Trades	Win Rate	Total PnL/Lot	Avg PnL/Lot
30	call	2.000	0.550	0.350	0.650	29	55.2%	$492	$17
30	call	2.000	0.550	0.350	0.650	29	55.2%	$492	$17
30	call	2.000	0.550	0.350	0.650	29	55.2%	$492	$17
30	call	2.000	0.500	0.350	0.650	30	53.3%	$416	$14
30	call	2.000	0.500	0.350	0.650	30	53.3%	$416	$14
30	call	2.000	0.500	0.350	0.650	30	53.3%	$416	$14
30	call	1.000	0.550	0.350	0.650	45	51.1%	$348	$8
30	call	1.000	0.550	0.350	0.650	45	51.1%	$348	$8
30	call	1.000	0.550	0.350	0.650	45	51.1%	$348	$8
30	call	1.000	0.500	0.350	0.650	46	50.0%	$307	$7
30	call	1.000	0.500	0.350	0.650	46	50.0%	$307	$7
30	call	1.000	0.500	0.350	0.650	46	50.0%	$307	$7
30	call	2.000	0.500	0.350	0.650	17	52.9%	$243	$14
30	call	2.000	0.550	0.350	0.650	17	52.9%	$243	$14
30	call	2.000	0.600	0.350	0.650	17	52.9%	$243	$14
30	call	2.000	0.650	0.350	0.650	17	52.9%	$243	$14
30	call	2.000	0.700	0.350	0.650	16	50.0%	$224	$14
30	call	2.000	0.700	0.350	0.650	16	50.0%	$224	$14
30	call	2.000	0.700	0.350	0.650	16	50.0%	$224	$14
30	call	2.000	0.700	0.350	0.650	16	50.0%	$224	$14

Close Decay Check

For the top exploratory pocket, the test marks the same 30-minute $2-wide call debit spread at several exit times. The actual path improves near the close, but the frozen-spot test shows that pure time decay is negative for this long debit spread.

Exit Min	Trades	Win Rate	Total PnL/Lot	Avg PnL/Lot	Exit Norm	Frozen Spot PnL
60.000	29.000	58.6%	$51	$2	41.7%	$-1
120.000	29.000	48.3%	$-58	$-2	39.8%	$-3
180.000	29.000	69.0%	$200	$7	44.2%	$-5
240.000	29.000	65.5%	$253	$9	45.2%	$-10
300.000	29.000	58.6%	$236	$8	44.9%	$-17
360.000	29.000	55.2%	$183	$6	44.0%	$-36
380.000	29.000	58.6%	$374	$13	47.3%	$-56
390.000	29.000	55.2%	$492	$17	49.3%	$-80

Real Option Bar Validation

This re-runs the credit-spread tests with Alpaca historical QQQ 0DTE option minute trade bars for entry marks, then intrinsic settlement at the close. This is closer to reality than Black-Scholes, but still not a bid/ask quote simulation.

Sample	Requested	Validated	Coverage	Real Win	Avg/Lot	Portfolio PnL	Conservative PnL
split_test	10	10	100.0%	100.0%	$47	$944	$847
fixed_60m_w5_p045_c010_030	57	56	98.2%	87.5%	$8	$844	$286