CuteMarkets Intraday Option Strategies

cuteoptionstrats packages one curated intraday options strategy on top of cutebacktests. The public surface is intentionally narrow: one named model (c36_quality), reproducible backtests, and explicit assumptions for signal generation, option selection, and out-of-sample evaluation.

The default model is the public alias for c36_vwap_mr_option_native_quality_v1, a quote-aware 0-2DTE VWAP mean-reversion options model for liquid U.S. index ETFs. This repository does not re-implement the full backtesting engine. It pins one research branch, exposes a small CLI and Python API, and documents the model and evaluation assumptions explicitly so developers can inspect an intraday options strategy without cloning the private research stack.

Quick links:

• Read docs
• Get API key
• Explore cutebacktests
• Explore cutemarkets-python

NOTE: This is not our best model, but a solid one.

Figure: trade-level ROIC_peak and drawdown from a completed out-of-sample server2 run of c36_vwap_mr_option_native_quality_v1, using the realized OOS trade window from 2025-10-07 to 2026-03-20.

Model In This Repository

• Default model id: c36_quality
• Backing profile: c36_vwap_mr_option_native_quality_v1
• Signal family: VWAP residual z-score mean reversion
• Instrument expression: single-leg options
• DTE window: 0-2DTE, target 1DTE
• Default universe: SPY, QQQ
• Runtime dependency: cutebacktests

Explore Examples And Artifacts

• examples/run_c36_quality.py
• examples/export_trade_log.py
• examples/inspect_model_metadata.py
• docs/assets/c36-quality-example-trades.csv for the public export shape

Why This Repo Exists Instead Of Publishing Every Strategy

The purpose of this repository is to publish one concrete model surface with explicit assumptions, not to release the full internal strategy set. That makes the public package more useful for developers because they can inspect one complete model card, one CLI, and one stable API without sorting through a large private research archive.

It also keeps the public contract honest. This repo is meant to show how CuteMarkets thinks about model packaging, quote-aware options backtesting, and out-of-sample reporting. It is not meant to imply that every stronger internal strategy should be public.

Where c36 Is Implemented

This repository selects and runs c36. The strategy definition itself lives in cutebacktests.

Layer	Location
Local default alias and model metadata	src/cuteoptionstrats/models.py
Local backtester construction	src/cuteoptionstrats/runtime.py
Local CLI	src/cuteoptionstrats/cli.py
Profile alias lookup (c36_quality)	cutebacktests: opening_range_profiles_registry.py
Exact profile definition (c36_vwap_mr_option_native_quality_v1)	cutebacktests: opening_range_profiles.py
Signal construction and exit resolution	cutebacktests: mean_reversion_intraday.py
Generic options backtester and contract/fill engine	cutebacktests: intraday_options.py

Operationally, cuteoptionstrats is the opinionated entrypoint. cutebacktests is the source of truth for the profile, signal engine, option selection, quote handling, and persistence.

Install

python3 -m venv .venv
source .venv/bin/activate
python -m pip install -e '.[dev]'

Creating The Data Directory And DuckDB Path

cuteoptionstrats accepts repo-scoped path variables and maps them to the underlying cutebacktests runtime:

• CUTEOPTIONSTRATS_DATA_DIR
• CUTEOPTIONSTRATS_DB_PATH

Recommended setup:

mkdir -p data

cat > .env <<'EOF'
CUTEOPTIONSTRATS_DATA_DIR=data
CUTEOPTIONSTRATS_DB_PATH=data/cuteoptionstrats.duckdb

CUTEMARKETS_API_KEY=your_cutemarkets_key
CUTEMARKETS_BASE_URL=https://api.cutemarkets.com
EOF

CuteMarkets is the default API path for this repository. Auxiliary provider support remains available in the runtime, but it is opt-in rather than part of the recommended public setup.

What happens when you run a backtest:

1. load_settings(".env") reads .env and maps CUTEOPTIONSTRATS_* paths onto the cutebacktests runtime settings.
2. Settings.from_env(...) creates the data directory and the parent directory of the database path if they do not exist.
3. DataStore(settings.db_path) opens DuckDB and initializes the schema on first writable access.

This means you do not need to pre-create the DuckDB file manually. Creating the directory is enough. The database file is created on first run.

What The Data Directory Contains

At minimum, the directory will contain the DuckDB file you configured, for example:

data/
└── cuteoptionstrats.duckdb

Depending on your workflow, the database accumulates:

• cached stock bars: stock_bars, stock_daily_bars, stock_bar_coverage
• cached option market data: option_chain, option_bars, option_quotes
• cached contract lookups and selection preferences: option_contract_cache, option_contract_list_cache, option_contract_universe_cache, option_contract_fetch_preference
• generated research artifacts written into the DB: signals, backtest_trades, paper_orders

The current public schema is defined in cutebacktests/src/cutebacktests/storage.py.

Practical guidance:

• use one DB per experiment branch if you want isolation
• keep the DB on a local SSD path if you plan to cache large options histories
• do not share one writable DuckDB across unrelated concurrent research jobs

Python Example

from datetime import datetime

from cuteoptionstrats import build_default_config, build_default_model, load_settings
from cuteoptionstrats.runtime import build_backtester
from cutebacktests.storage import DataStore

settings = load_settings(".env")
model = build_default_model()
store = DataStore(settings.db_path)

try:
    backtester = build_backtester(settings=settings, store=store)
    result = backtester.run(
        build_default_config(
            start=datetime(2025, 1, 1),
            end=datetime(2025, 1, 31),
            ticker="SPY",
            return_trade_log=True,
        )
    )
    print(model.profile_name, result["trades"])
finally:
    store.close()

CLI

Show the default model metadata:

python -m cuteoptionstrats.cli show-model

Run a backtest with the default model:

python -m cuteoptionstrats.cli run-backtest \
  --ticker SPY \
  --start 2025-01-01 \
  --end 2025-01-31

Run the default universe in one shot:

python -m cuteoptionstrats.cli run-default-universe \
  --start 2025-01-01 \
  --end 2025-01-31

Inspect the fully resolved config payload:

python -m cuteoptionstrats.cli show-effective-config

c36 Model Specification

Profile Lineage

c36_quality is not a standalone handwritten strategy. It is built by composition:

1. c4_mr_vwap_zscore_v2
2. c4_mr_vwap_zscore_v2_fast
3. c18_vwap_mr_balance_v1
4. c18_vwap_mr_quality_v1
5. c36_vwap_mr_option_native_quality_v1

That lineage matters because the final profile inherits both the original VWAP z-score logic and the later option-native execution overlay.

Active Parameters

The final c36_quality configuration inherits the following effective settings from the profile chain.

Parameter	Value	Role
strategy_variant	mr_vwap_zscore_v2	use the VWAP residual z-score engine
opening_range_minutes	5	first regular-session bars used only to define early context
entry_start_time	09:35	earliest signal evaluation time
entry_cutoff_time	12:00	no new entries after noon ET
exit_time	15:55	hard intraday exit cutoff
allow_long	True	long mean-reversion entries allowed
allow_short	True	short mean-reversion entries allowed
require_breakout_open_inside_range	False	no ORB-style inside-range gate
require_relative_volume	True	require elevated opening activity
relative_volume_min	1.05	minimum relative opening volume
mr_zscore_window	20	rolling residual standard deviation window
mr_zscore_entry	1.35	excursion threshold
mr_zscore_reentry	0.65	reclaim threshold after excursion
mr_zscore_stop	2.15	stop envelope in z-score space
mr_zscore_target	0.15	z-score target used for profit-taking
mr_take_profit_mode	zscore	target expressed off VWAP residual space
mr_require_reversal_candle	True	signal bar must close in the reversal direction
exit_on_opposite_candle	True	position can exit on opposite-color candle
opposite_candle_min_hold_minutes	3	minimum hold before opposite-candle exit
max_hold_minutes	25	strict time budget
mr_sigma_min_pct	0.0005	reject sessions with too little residual volatility
mr_sigma_max_pct	0.0160	reject sessions with too much residual volatility
mr_vwap_slope_lookback	3	slope lookback for VWAP trend filter
mr_vwap_slope_max_pct	0.0020	reject strongly trending VWAP states
mr_stop_buffer_or_mult	0.15	stop buffer in opening-range units
option_min_dte	0	same-day options allowed
option_target_dte	1	next-day expiry is preferred
option_max_dte	2	maximum expiry horizon
require_option_microstructure_filter	True	option-side quality gate enabled
option_min_entry_volume	25	minimum option bar volume
option_min_entry_price	0.60	avoid extremely cheap contracts
option_max_entry_bar_range_pct	0.35	reject highly unstable entry bars
option_min_quote_coverage_pct	0.35	quote coverage floor
option_min_chain_coverage_pct	0.35	option chain coverage floor
option_structure_mode	single_leg	no spread structure by default
option_structure_filter_enabled	False	no extra structure filter layer on top

Exact Entry Logic

The signal engine is implemented in cutebacktests/src/cutebacktests/strategies/mean_reversion_intraday.py.

For each session:

1. Compute the opening range from the first 5 bars.
2. Compute a running intraday VWAP from session bars.
3. Compute residuals as close - VWAP.
4. Compute a rolling standard deviation of those residuals over a 20-bar window.
5. Convert residuals into a VWAP residual z-score: z_t = (close_t - VWAP_t) / sigma_t.

The model then scans bars from 09:35 through 12:00 ET and rejects the bar unless all active filters pass:

• relative opening volume must be at least 1.05
• residual volatility must satisfy 0.0005 <= sigma / close <= 0.016
• absolute VWAP slope over the last 3 bars must be at most 0.002
• the signal bar must be a reversal candle in the trade direction
• the setup must still allow a valid stop and next-bar entry

The long setup is:

long_signal =
    (
      z_prev <= -1.35 and z_now >= -0.65
    )
    or
    (
      low <= VWAP - 1.35 * sigma
      and close >= VWAP - 0.65 * sigma
    )

plus a bullish reversal candle (close > open).

The short setup is symmetric:

short_signal =
    (
      z_prev >= 1.35 and z_now <= 0.65
    )
    or
    (
      high >= VWAP + 1.35 * sigma
      and close <= VWAP + 0.65 * sigma
    )

plus a bearish reversal candle (close < open).

If a signal is accepted, the trade enters on the next bar open, not on the signal bar. This is an important causal assumption. The signal bar is used for detection. The fill occurs one bar later.

Exact Stop And Target Logic

For a long trade:

stop_core = min(signal_bar_low, previous_bar_low, VWAP - 2.15 * sigma)
stop = stop_core - 0.15 * opening_range_width
target = VWAP + 0.15 * sigma

For a short trade:

stop_core = max(signal_bar_high, previous_bar_high, VWAP + 2.15 * sigma)
stop = stop_core + 0.15 * opening_range_width
target = VWAP - 0.15 * sigma

If the z-score target would sit on the wrong side of the entry price, the code normalizes the target to a minimum economically meaningful distance from entry using the larger of:

• 0.4 * risk_per_share
• 0.5 * sigma
• 0.2 * opening_range_width

Exact Exit Logic

After entry, c36 exits on the earliest triggered condition:

1. stop loss
2. z-score target hit (exit_reason = mr_target_zscore)
3. opposite-color candle after the 3 minute minimum hold
4. max hold time of 25 minutes
5. time-based exit at or after 15:55
6. final fallback at the session close if none of the above fired

Features that are present in the generic engine but not active in the c36_quality profile by default:

• break-even trailing via break_even_trigger_rr
• early-fail exit via early_fail_minutes
• time-to-work veto
• adaptive z-score thresholds
• macro-release blocking
• OR-width filtering
• trend-alignment filtering

Option Expression And Contract Assumptions

c36_quality is the option-native version of the c18 signal family. The signal is produced on the underlying, then monetized through the generic cutebacktests options runner.

The profile-level option assumptions are:

• single-leg options, not spreads
• target expiry around 1DTE
• allowed expiry window 0DTE to 2DTE
• quote-aware option execution
• option microstructure filter enabled
• minimum option entry volume 25
• minimum option entry premium 0.60
• maximum entry-bar option range 35%
• minimum quote coverage 35%
• minimum chain coverage 35%

The full contract-ranking and fill engine lives in cutebacktests/src/cutebacktests/backtest/intraday_options.py. This repository does not override that machinery.

Assumptions And Limitations

The model should be interpreted with the following assumptions in mind:

• it is an intraday, same-session strategy; no overnight holding is intended
• entries are next-bar causal, not same-bar discretionary fills
• VWAP and z-score statistics are recomputed inside the session from the available session bars
• the profile assumes liquid underlying instruments such as SPY and QQQ
• option selection quality matters materially because the trade horizon is short and the option leg is close-dated
• c36 is a sparse model by design; that sparsity is part of the research result, not an implementation accident

Evaluation Metrics

The research stack uses a portfolio-selection vocabulary rather than a single headline return number. The most important metrics surfaced in the public artifacts are:

Metric	Meaning in this research stack
Sharpe_yearly	annualized Sharpe ratio computed from realized daily returns when daily returns are available
Sortino_yearly	annualized Sortino ratio computed from realized daily returns
DSR	Deflated Sharpe Ratio style probability adjustment that discounts the reported Sharpe for multiple testing and non-Gaussian return shape; implemented from fold-level returns, strategy count, skew, and kurtosis
PBO	Probability of Backtest Overfitting, estimated via CSCV-style train/test split enumeration across fold results; lower is better
ROIC_peak	out-of-sample PnL divided by peak capital deployed, where peak capital deployed is the maximum capital committed at one time/trade in the evaluated stream
CAGR_ROIC_peak	compound annualization of ROIC_peak over the out-of-sample evaluation span
Trades / week	trade density; used because a profitable but inactive model may still fail its intended portfolio role

ROIC_peak Definition

ROIC_peak is not the same thing as total equity return.

In the research code, it is computed as:

ROIC_peak = out_of_sample_sum_pnl / out_of_sample_peak_capital_deployed

where out_of_sample_peak_capital_deployed is the largest deployed capital footprint observed in the evaluated stream. The purpose of this metric is capital efficiency: how much out-of-sample PnL the strategy generated relative to the maximum capital it had to tie up at peak usage.

How DSR And PBO Are Calculated Here

PBO is computed in the research stack with CSCV-style splits over fold-level out-of-sample results. In practice, it asks how often the strategy that wins on the training side does not hold up well on the test side. Lower values indicate less evidence of selection-driven overfitting.

DSR is computed from fold-level returns per strategy. The implementation adjusts raw Sharpe for:

• the number of active strategy trials
• sample size
• skew
• kurtosis

The output used in ranking is the best profile-level dsr value in [0, 1], not raw Sharpe itself.

The metric implementations currently live in the internal research code at trading2/src/capcopy/research/orb_conviction.py. For public readers, the important point is methodological: the stack explicitly penalizes multiple testing and overfitting rather than ranking only by raw return.

Best Published Results So Far

The public/local artifact set currently exposes the following strategy-level results explicitly.

Model	Published evidence	Interpretation
c66_opening_compression_option_native_short_balance_dte35_v1	base OOS return 19.18%, stress-medium 16.70%, stress-harsh 15.56%, 76 OOS trades in all three cases	strongest current internal baseline
c36_vwap_mr_option_native_quality_v1	+16004 PnL, 15 trades, DSR 0.6400, failed only trades_per_week_ok	real but sparse runner-up; this repository packages this branch
c36_vwap_mr_option_native_opportunity_v1	85 trades, +2987 PnL	density improved, quality decayed

Tests

python -m pytest -q