xpertsystems/enr006-sample

ENR006 — Synthetic Wholesale Energy Market Trading Dataset (Sample Preview)

XpertSystems.ai | Synthetic Data Factory | Energy & Climate Vertical

A six-table wholesale energy market trading dataset spanning the full trading lifecycle: hourly Day-Ahead LMPs (energy + congestion + loss three-part decomposition), futures / forwards / swaps / CfDs with options Greeks, six ancillary services markets (REG_UP, REG_DOWN, SPINNING_RESERVE, NON_SPIN_RESERVE, BLACK_START, VOLTAGE_SUPPORT), market clearing with imports/exports and energy balance, OTC bilateral PPAs with credit exposure, and per-trade execution analytics with Basel III coherent risk metrics (VaR-95, VaR-99, CVaR-95, Sharpe). Calibrated benchmark-first against FERC Order 755/888/890, NERC reliability standards, ISDA Master Agreement, EEI Master Agreement, Basel III FRTB, Schwartz (1997) mean-reversion theory, and EIA/PJM/CAISO/ERCOT 2023 published LMP data.

This is the sample preview — 2 weeks (336 hours) of hourly DA market data + 500 futures + 300 bilateral + 2,000 trades + 1 week of ancillary services clearing (~13K total records). The full product covers a full annual cycle × 500 pricing nodes × 200 participants × 20K trades with pre-built scenario configs for price-spike events, high-renewable negative pricing, and capacity-crunch market stress.

---

Dataset summary

Table	Rows (sample)	What it contains
spot_price	1,680	Hourly DA LMP with three-part decomposition: lmp_total = energy + congestion + loss, plus system_lambda, peak/off-peak flags, weekend/holiday flag, price cap and negative price event flags
futures_contracts	500	FUTURES / FORWARD_OTC / SWAP / CfD contracts: tenors (DAY/WEEK/MONTH/QUARTER/CALENDAR_YEAR), forward curve, basis, contract price, notional, options Greeks (delta/gamma/vega/theta), MTM, settlement P&L
ancillary_services	~8,500	Hourly clearing for 6 services: clearing price, capacity awarded, performance score, mileage (REG_UP/DOWN), activation flag/duration, obligation, availability payment
market_clearing	336	DAM clearing: total cleared load/gen/imports/exports, energy balance (zero by construction), reserve margin, convergence flag, virtual bid volume + P&L, interchange schedule, market surplus, demand response cleared, capacity market price
bilateral_contracts	300	OTC PPAs: FIXED_PRICE / INDEXED / SHAPED / TOLLING structures, product type (FIRM / NONFIRM / UNIT_CONT / SYSTEM), volume, duration, fixed price, index reference, adder, total contract value, credit exposure, collateral posted, counterparty credit rating, EEI confirmation flag
trading_analytics	2,000	Per-trade execution: timestamp (ms-precision), trader / book, BUY/SELL direction, quantity, execution and market price, slippage, transaction cost, realized + unrealized P&L, VaR_95, VaR_99, CVaR_95, Sharpe ratio, max drawdown, position, hedge ratio, regulatory flag

All six tables are provided in both CSV and Parquet. They join on node_id, participant_id (= buyer_id / seller_id / trader_id / provider_id), book_id, and timestamp_utc.

---

Calibration sources

All ten validation metrics target named industry sources, not generator self-metrics:

• FERC Order 888 / 890 — Open Access Transmission Tariff, LMP three-part decomposition (energy + congestion + loss)
• FERC Order 755 — Pay-for-performance regulation (REG_UP / REG_DOWN clearing structure)
• FERC Ancillary Services Tariffs (PJM / CAISO / ERCOT 2023) — six ancillary product price ranges
• NERC TPL-001-5 — bulk system energy balance requirements
• NERC LOLP / IEEE Reliability Standards — reserve margin planning ranges (12-25% typical, 5-40% observed)
• ISDA Master Agreement — notional value definition for derivatives
• EEI Master Agreement — bilateral power transaction value calculation
• Basel III FRTB + Artzner et al. (1999) — coherent risk measure axioms (CVaR ≥ VaR, monotonicity in confidence level)
• EIA / PJM / CAISO / ERCOT 2023 — published wholesale hub LMP averages for cross-ISO calibration
• Schwartz (1997) / Lucia & Schwartz (2002) — mean-reverting commodity price model theory

---

Validation scorecard (seed = 42)

10/10 PASS · Grade A+ (100%) across all six canonical seeds (42, 7, 123, 2024, 99, 1).

#	Metric	Observed	Target	Tol	Type	Source
1	lmp_decomp_identity_normal_rows_rate	1.000	0.99	±0.01	FLOOR	FERC Order 888/890
2	market_clearing_energy_balance_zero_rate	1.000	0.99	±0.01	FLOOR	NERC TPL-001-5
3	var_coherence_rate	1.000	0.99	±0.01	FLOOR	Basel III / Artzner 1999
4	futures_notional_identity_rate	1.000	0.99	±0.01	FLOOR	ISDA Master Agreement
5	bilateral_total_value_identity_rate	1.000	0.99	±0.01	FLOOR	EEI Master Agreement
6	ancillary_clearing_prices_in_iso_bounds_rate	1.000	0.99	±0.01	FLOOR	FERC AS tariffs
7	reserve_margin_in_industry_range_rate	1.000	0.95	±0.05	FLOOR	NERC LOLP / IEEE
8	lmp_mean_usd_per_mwh_in_iso_band	40.26	45.0	±20.0	two-sided	EIA/PJM/CAISO/ERCOT 2023
9	reg_up_clearing_price_mean_usd_per_mw_hr	15.56	15.0	±5.0	two-sided	FERC Order 755 / PJM REG-UP
10	spot_price_in_iso_floor_cap_bounds_rate	1.000	0.99	±0.01	FLOOR	FERC/PJM Tariff

---

Schema highlights

spot_price (1,680 rows × 13 cols)

node_id, timestamp_utc, settlement_type (DAM), lmp_total_usd_per_mwh, lmp_energy_usd_per_mwh, lmp_congestion_usd_per_mwh, lmp_loss_usd_per_mwh, system_lambda_usd_per_mwh, price_hub, hour_ending, peak_offpeak_flag (ON_PEAK / OFF_PEAK), weekend_holiday_flag, price_cap_flag, negative_price_flag.

futures_contracts (500 rows × 23 cols)

contract_id, contract_type (FUTURES / FORWARD_OTC / SWAP / CfD), underlying_hub, node_id, tenor (DAY / WEEK / MONTH / QUARTER / CALENDAR_YEAR), delivery_start_utc, delivery_end_utc, trade_date_utc, contract_price_usd_per_mwh, forward_curve_usd_per_mwh, basis_usd_per_mwh, contract_quantity_mwh, notional_value_usd, buyer_id, seller_id, trader_book, mark_to_market_usd_per_mwh, implied_vol_pct, delta, gamma, vega, theta, settlement_price_usd_per_mwh, settlement_gain_loss_usd.

ancillary_services (~8,500 rows × 13 cols)

ancillary_id, timestamp_utc, service_type (REG_UP / REG_DOWN / SPINNING_RESERVE / NON_SPIN_RESERVE / BLACK_START / VOLTAGE_SUPPORT), clearing_price_usd_per_mw_hr, capacity_awarded_mw, performance_score_pct, mileage_mw, mileage_payment_usd, activation_flag, activation_duration_min, provider_id, zone_id, obligation_mw, availability_payment_usd.

market_clearing (336 rows × 17 cols)

clearing_id, timestamp_utc, market_type (DAM), clearing_timestamp_utc, total_cleared_load_mw, total_cleared_gen_mw, total_cleared_imports_mw, total_cleared_exports_mw, energy_balance_mw, reserve_margin_pct, convergence_flag, virtual_bid_volume_mwh, virtual_bid_pnl_usd, interchange_schedule_mw, market_surplus_usd, demand_response_cleared_mw, capacity_market_price_usd_per_mw_day, system_lambda_usd_per_mwh.

bilateral_contracts (300 rows × 18 cols)

bilateral_id, trade_date_utc, contract_structure (FIXED_PRICE / INDEXED / SHAPED / TOLLING), buyer_id, seller_id, delivery_point, node_id, product_type (FIRM / NONFIRM / UNIT_CONT / SYSTEM), volume_mw, duration_months, fixed_price_usd_per_mwh, index_reference, adder_usd_per_mwh, total_contract_value_usd, credit_exposure_usd, collateral_posted_usd, counterparty_credit_rating (AAA / AA / A / BBB / BB / B / CCC), eei_confirmation_flag.

trading_analytics (2,000 rows × 19 cols)

trade_id, execution_timestamp_utc (ms precision), trader_id, book_id (BOOK_01..BOOK_20), trade_direction (BUY / SELL), trade_quantity_mwh, execution_price_usd_per_mwh, market_price_usd_per_mwh, slippage_usd_per_mwh, transaction_cost_usd, realized_pnl_usd, unrealized_pnl_usd, var_95_usd, var_99_usd, cvar_95_usd, sharpe_ratio, max_drawdown_usd, position_mw, hedge_ratio_pct, regulatory_flag.

---

Suggested use cases

• Day-ahead LMP forecasting — train regressors / LSTMs for lmp_total_usd_per_mwh from time-of-day, day-of-year, peak/off-peak, and historical lag features
• Three-part LMP decomposition modeling — predict lmp_congestion_usd_per_mwh and lmp_loss_usd_per_mwh separately from topology / loading proxies for FTR / CRR markets
• Price spike detection — anomaly classifier for price_cap_flag and negative_price_flag from system_lambda, peak_offpeak_flag, and weather proxies (pair with ENR-002 weather data)
• Futures forward curve modeling — fit yield-curve / forward-curve structures from tenor, delivery_start_utc, contract_price, forward_curve triples
• Options Greeks calibration — train Black-76 / spread option models on implied_vol_pct, delta, gamma, vega, theta for options-on-futures pricing
• Ancillary services co-optimization — joint price models for energy + AS clearing across 6 services
• Bilateral PPA pricing — model fixed_price_usd_per_mwh as a function of volume_mw, duration_months, counterparty_credit_rating, and index_reference; useful for term-sheet automation
• Credit risk / counterparty exposure — train default probability models from counterparty_credit_rating joined with credit_exposure_usd and collateral_posted_usd
• VaR backtesting — use the included VaR_95 / VaR_99 / CVaR_95 columns as benchmarks for new ML-driven VaR models; check coherence axioms
• Slippage modeling — predict slippage_usd_per_mwh from quantity, market_price, and time-of-day; useful for transaction cost analysis
• Virtual bidding (INC/DEC) strategies — train signal models from virtual_bid_volume_mwh and virtual_bid_pnl_usd joined with LMP changes between DAM and RTM
• Regulatory flag detection — multi-class for regulatory_flag from trade-level signals (quantity, slippage, market deviation); useful for market surveillance / spoofing detection
• Capacity market clearing modeling — predict capacity_market_price_usd_per_mw_day from reserve_margin_pct and load growth trends
• Demand response clearing — model demand_response_cleared_mw from LMP and load shape signals

---

Loading examples

from datasets import load_dataset

spot = load_dataset("xpertsystems/enr006-sample", "spot_price", split="train")
futures = load_dataset("xpertsystems/enr006-sample", "futures_contracts", split="train")
print(spot.shape, futures.shape)

import pandas as pd
from huggingface_hub import hf_hub_download

spot = pd.read_parquet(hf_hub_download(
    "xpertsystems/enr006-sample", "enr006_spot_price.parquet",
    repo_type="dataset",
))

# LMP three-part decomposition: verify the identity on non-cap, non-negative-price rows
normal = spot[(spot["price_cap_flag"] == 0) & (spot["negative_price_flag"] == 0)]
residual = (
    normal["lmp_total_usd_per_mwh"]
    - normal["lmp_energy_usd_per_mwh"]
    - normal["lmp_congestion_usd_per_mwh"]
    - normal["lmp_loss_usd_per_mwh"]
).abs()
print(f"Max decomp residual on normal rows: {residual.max():.6f}")
print(f"Mean decomp residual: {residual.mean():.6f}")

# Build a simple forward curve from futures
import pandas as pd
from huggingface_hub import hf_hub_download

futures = pd.read_parquet(hf_hub_download(
    "xpertsystems/enr006-sample", "enr006_futures_contracts.parquet",
    repo_type="dataset",
))

# Average price by tenor
print(futures.groupby("tenor")["contract_price_usd_per_mwh"].agg(["mean", "std", "count"]))

# Trader P&L attribution
import pandas as pd
from huggingface_hub import hf_hub_download

trd = pd.read_parquet(hf_hub_download(
    "xpertsystems/enr006-sample", "enr006_trading_analytics.parquet",
    repo_type="dataset",
))

book_pnl = trd.groupby("book_id").agg(
    realized=("realized_pnl_usd", "sum"),
    n_trades=("trade_id", "count"),
    avg_var95=("var_95_usd", "mean"),
).round(2).sort_values("realized", ascending=False)
print(book_pnl.head(10))

# Validate VaR coherence (Basel III requirement)
import pandas as pd
from huggingface_hub import hf_hub_download

trd = pd.read_parquet(hf_hub_download(
    "xpertsystems/enr006-sample", "enr006_trading_analytics.parquet",
    repo_type="dataset",
))

var99_ge_var95 = (trd["var_99_usd"] >= trd["var_95_usd"]).mean()
cvar95_ge_var95 = (trd["cvar_95_usd"] >= trd["var_95_usd"]).mean()
print(f"VaR_99 >= VaR_95: {var99_ge_var95*100:.2f}%")
print(f"CVaR_95 >= VaR_95: {cvar95_ge_var95*100:.2f}%")

---

Limitations and honest disclosures

This sample is calibrated for structural fidelity, not bit-exact reproduction of any specific ISO's settlement archive. Specifically:

• Spot prices cover only 5 pricing nodes even when n_pricing_nodes is set higher — the generator hardcodes node_ids[:5] in its main flow (line 821). This is intentional sample-mode behavior; the full product pipeline scales to 500+ pricing nodes via the per-node-batch design.
• Ancillary services covers only the first 168 timestamps (one week via the generator's timestamps_da[:168] slice on line 823). For hours_da=336 in this sample, ancillary spans week 1 only; spot, futures, market clearing, and trading span the full 2-week window.
• LMP three-part decomposition is broken by design on (a) rows where lmp_total is clamped to the ISO price cap or floor, and (b) rows where negative_price_flag=1 (negative-price override). The wrapper validates the decomposition on NORMAL rows only (cap_flag=0 AND negative_price_flag=0). For research that requires the full identity to hold, mask out the special-case rows or use the lmp_energy + lmp_congestion + lmp_loss sum directly.
• Price spike rate and negative price rate are sample-scale unstable. At 1,680 spot rows, the generator's Poisson(0.02) spike arrivals and the conjunction random < 0.04 AND system_lambda < 0.3*base_lmp for negative prices fire too rarely to validate against the generator's designed 2-10% spike / 1-8% negative-price targets. The full annual product matches those targets at scale. For tail-event ML, use the full product or the pre-built scenario configs.
• system_lambda AR(1) coefficient is HIGHLY seed-dependent at sample scale (observed range 0.20-0.90 across 6 seeds). The underlying mean-reverting process has θ=0.12/hr → asymptotic AR(1) ≈ 0.88, but spike events at small samples distort the lag-1 correlation. We validate system_lambda falling within the ISO floor/cap bounds instead of the AR(1) coefficient. For mean-reversion analysis, use the full annual product or fit a state-space model.
• The generator's run_benchmarks reports "Grade: A+" misleadingly. Its all_passed flag is only updated by check_list (line 598-603) which isn't actually invoked for any test in this version — so all_passed=True regardless of module-level pass/fail flags. This wrapper provides genuine industry-anchored validation via the scorecard above.
• Forward curve uses a simplified seasonal + linear-risk-premium shape (fwd_curve = mean(system_lambda) * seasonal_adj + risk_premium). Real forward curves include calendar-spread structure, weather-stochastic vol, and counterparty-specific basis that the generator does not model.
• Options Greeks fire on only ~15% of contracts (FUTURES + CfD types with 30% optionality probability). The remaining 85% have all Greeks set to zero. Filter implied_vol_pct > 0 to extract the options- bearing subset before training Greek-prediction models.
• negative_price_flag = 1 row prices use -rng.exponential(20) — i.e., a magnitude draw, NOT a structural reason like solar oversupply or congestion island. Use the flag as a label, not a causal driver.
• market_surplus_usd, virtual_bid_pnl_usd, capacity_market_price are independent random draws, not computed from underlying market dynamics. Treat as auxiliary fields for model-feature space, not as ground-truth market clearing outputs.
• Credit ratings are sampled with a designed distribution [5%, 10%, 20%, 30%, 20%, 10%, 5%] for [AAA, AA, A, BBB, BB, B, CCC] — IG-skewed but not anchored to any specific counterparty pool. Use as a categorical feature; don't infer real-world default probabilities directly.
• All trades sampled from participant_ids uniformly; trade pairings (buyer / seller) can occasionally match the same participant for both sides at small sample scale. For market-surveillance ML, filter buyer_id ≠ seller_id.
• hours_da is hourly cadence only — no 5-min real-time market data in the sample. The full product includes both DAM and RTM at 5-min resolution via intervals_rt_per_hour=12.

The full ENR006 product addresses these by full annual coverage, all 500+ pricing nodes, calibrated forward curves, RTM 5-min interval settlement, and pre-built scenario configs (price_spike_event, high_renewable_negative_prices, capacity_crunch, standard_annual) — contact us for the licensed commercial release.

---

Companion datasets in the Energy & Climate vertical

• ENR-001 — Synthetic Power Grid Operations Dataset (transmission bus telemetry, line flows, generation dispatch, frequency, contingency)
• ENR-002 — Synthetic Renewable Energy Generation Dataset (utility-scale solar/wind/hybrid SCADA, weather, forecast, PCC, BESS)
• ENR-003 — Synthetic Electricity Demand & Load Forecasting Dataset (zone-level demand, multi-horizon forecasts, peak events, EV/DER, TOU)
• ENR-004 — Synthetic Upstream Oil & Gas Production Dataset (well-level production, decline curves, PVT, commodity prices, Subpart W methane)
• ENR-005 — Synthetic Smart Grid Dataset (AMI, DER, OpenADR, feeder power flow, grid edge analytics)
• ENR-006 — Synthetic Wholesale Energy Market Trading Dataset (you are here) — the market/trading complement to ENR-001's physical-grid view: spot price formation, derivatives, ancillary services, bilateral PPAs, and trading risk

Use ENR-001 + ENR-003 + ENR-006 together for full physical-grid + load-forecast + market-clearing ML workflows; combine with ENR-002 + ENR-005 to add renewables and distribution-edge in the same modeling stack.

For subsurface companion data (seismic, well logs, reservoir simulation, geological formations), see the OIL series (OIL-001 through OIL-004) in our Oil & Gas vertical.

For the broader catalog:

• Materials & Energy — MAT-001
• Insurance & Risk — 10 SKUs
• Cybersecurity — 11 SKUs

---

Citation

@dataset{xpertsystems_enr006_sample_2026,
  author       = {XpertSystems.ai},
  title        = {ENR006 Synthetic Wholesale Energy Market Trading Dataset (Sample Preview)},
  year         = 2026,
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/xpertsystems/enr006-sample}
}

---

Contact

• Web: https://xpertsystems.ai
• Email: pradeep@xpertsystems.ai
• Full product catalog: Cybersecurity, Insurance & Risk, Materials & Energy, Oil & Gas, Energy & Climate, and more

Sample License: CC-BY-NC-4.0 (Creative Commons Attribution-NonCommercial 4.0) Full product License: Commercial — please contact for pricing.