Video summary

Ex-Citadel Quant on Trading the Most Asymmetric Market - Neel Somani

Main summary

Key takeaways

Finance

Finance-focused summary (power & gas market trading, quant role, risk, and strategy)

Market structure & why power trading is special

  • Power markets are highly opaque and asymmetric.
  • In power, a single cold week can move prices up to ~100x, and being on the wrong side can be catastrophic.

  • Always-long power is described as a structurally losing strategy: the market contains positive skew events (rare extreme spikes). If you’re always long, you’re effectively paying for that skew without being able to hedge it—so on average you lose money unless you have a specific regional thesis.

  • The core edge in power trading is often congestion:

    • Congestion arises when transmission lines hit capacity (“ratings”).
    • When power can’t physically move from region A to B (or is physically lost), regional prices diverge, creating spreads that traders can target.

Quant researcher / model-building in hedge funds (how trades are produced)

Different hedge fund types imply different quant researcher (QR) roles:

  • Fully automated systematic funds (e.g., Two Sigma / “Two Sigma dehaw”)

    • QR builds models that output relative scores across assets
    • Scores convert to positions, and an execution system places trades

  • Discretionary shops (described as Citadel commodities; “not fully automated”)

    • QR builds models used by traders
    • Model outputs a price estimate (or inputs helping traders judge “what’s already priced in”)
    • Discretionary traders blend model output with uncertainty and other models

  • Desk analysts embedded on trading desks (power/gas)

    • More news-following / granular event monitoring
    • Example given: tracking generator start dates pushed back by a month at fine granularity

P&L attachment

  • QR is aware of desk P&L and attends PM meetings; junior quant slope is limited.
  • Seniority increases compensation via “slope”, and potentially negotiating for a % of P&L once placing trades.

Fundamental drivers and “granularity” in power & gas

  • “Table stakes” fundamentals don’t change, but top analysts track very granular operational details, e.g.:
    • specific plants
    • latest outages
    • timing of operational changes
  • The modeling foundation is described as “micro econ” + grid optimization / ISO rules.
    • Example: California ISO (CAISO)
    • Pricing computed from daily auctions and Kaiso rules, with reliability constraints partially inferred.

How power prices are modeled (frameworks described)

Two trading styles are contrasted: hub trading (easier) vs congestion/basis trading (harder).

Hub trading (directional on average regional price) — step framework

  1. Model demand using weather to estimate electricity demand (hot/cold drives heating/AC usage).
  2. Infer which generation units are needed
    • Low demand: renewables cover much of load
    • High demand: run less efficient units → prices rise
  3. Estimate price for a specific time period.
  4. Compare model estimate vs market price
    • If mispricing persists, determine whether the model is wrong or whether there is tradeable edge.

Congestion / basis trading (spreads between regions)

  • Edge often comes from expecting future spreads between locations.
  • Requires:
    • Generation vs demand estimates at each region (local capacity + imports/exports)
    • A global view of nearby regions, since imports/exports elsewhere affect whether lines become congested
  • Physical intuition:
    • Power lines have thermal/mechanical limits (expansion → sag → risk).
    • When line limits are reached, capacity can’t be used → price divergence.

Order placement / collateral requirements / access constraints

  • Trading involves onboarding with the ISO (example: California ISO onboarding).
  • Because power prices can spike extremely high—especially when short—traders may post seven figures of collateral to trade seriously.
  • Retail/smaller funds may struggle to access power trading due to capital intensity.

Natural gas ↔ power linkage

  • Power production often requires natural gas (and sometimes coal) as dispatchable thermal sources.
  • Coal is also mentioned as relevant in the Midwest (MISO referenced as “Mso”).
  • Core idea: the marginal “last unit” (gas/coal) often sets price → so gas fundamentals are crucial for power.

Why hedge funds trade power & gas more than oil

  • Oil is not generally used to produce power (except in extreme scenarios, e.g., some regions during severe winter shortages).
  • Petroleum prices are more exposed to global geopolitics (example: Strait of Hormuz closure affecting global supply/prices).
  • US natural gas is described as less globally sensitive domestically due to LNG exports—foreign price moves don’t transmit 1:1 to domestic gas pricing.

Risk management & the dominant failure mode

Biggest risk factor

  • Failure to account for “skew events” (low probability, extreme outcomes) is cited as the #1 way traders lose money.

Scarcity risk mechanics (why shorting is dangerous)

  • When demand can’t be curtailed (e.g., hospitals, AC in heat, heating in cold), scarcity can drive extreme prices.
  • If the grid can’t meet demand:
    • blackouts/brownouts occur
    • prices can hit near the highest possible level
  • Short positions may earn risk premium, but can also blow up if the skew event occurs against you.

Concrete example trade & exit logic (Texas 2021 freeze)

  • Example window: March / February 2021 in Texas.
  • Narrative:
    • Overconfident weather model → trader goes short power (or related exposure).
    • As updated weather increases probability of a freak freeze, they may double down.
    • Shortly before the event, uncertainty implies:
      • ~30% chance of the event
      • If maximum price: up to ~$9,000 per MWh
      • Normal range: ~$50–$100 per MWh
      • Implied move: 10–20x (and potentially ~100x per earlier discussion)
    • The “good trader” closes to avoid blow-up.
    • After the freeze day arrives, the model realizes the event timing and prices settle.

What to model (risk/inputs checklist)

  • Weather (drives demand)
  • Outages (subtract unavailable supply)
  • Fuel costs
    • natural gas price sensitivity to temperature (heating demand)
    • local coal shortages/prices if relevant
  • Renewable availability (wind not windy / solar not sunny)
  • New plants online and overall supply adequacy

Where multi-asset funds focus (basis vs directional)

  • Basis trade definition: long one area, short another (a congestion bet between regions).
  • Directional: expect hub price or a specific node to rise/fall.
  • Heuristic:
    • Basis is “easier” to model (more naturally a congestion framework).
    • Many managers are delta neutral (both long/short).
    • Straight long a single zone is less common because modeling that zone implies modeling surrounding regions too.

Performance and strategy implications (explicit recommendations/cautions)

  • Caution: Don’t assume “always long” is viable in power; the skew structure makes it structurally losing.
  • Trading implication: Edge is more often found in congestion/basis than in pure directional hub bets, due to modeling tractability and variance.
  • Risk implication: Plan for skew events; if probability rises enough and max loss is extreme (e.g., $9,000/MWh vs $50–$100), risk controls lead to closing.

Macro / AI + commodities demand context (post-Citadel career)

  • He says AI is increasing power demand, feeding compute/data-center energy needs.
  • Mentions relevant inputs to AI/commodities/power:
    • data
    • compute
    • fundamental research (especially compute/power linkage)
  • Discusses a 6–12 month power demand planning problem: how to meet demand without locking into wrong capacity (and whether to lock in).

Temporary/dirtier generation (diesel) considerations

  • Diesel becomes more relevant depending on AI lab willingness to pay and project economics.
  • Example willingness-to-pay:
    • $150/MWh might be “too high”
    • $50/MWh is suggested as a more plausible range where financing a ~6-month commitment might be economical
  • Example costs/finance points:
    • ~$80m–$100m for a small plant (~50 MW implied)
    • interest rate example: ~20% YoY
    • production cost example: ~$50/MWh
  • Core point: economics hinge on the lab’s actual P&L willingness at specific $/MWh.

“Tab” theory on supply cycles

  • Referenced a theory: many gluts aren’t followed by shortages, but shortages are often followed by gluts because participants overbuild or rush supply and mistime response.

Disclosures / disclaimers

  • No explicit “financial advice” disclaimer appears in the subtitles provided.

Tickers / instruments / assets mentioned

  • Natural gas
  • Coal
  • Power (electricity markets)
  • Diesel (as temporary power generation)

(No specific equities/ETFs/bonds/tickers were mentioned.)

Key numbers / metrics called out

  • Collateral: “seven figures” for power trading seriousness
  • Power spike magnitude: ~100x from one cold week
  • Texas 2021 example:
    • Max price: ~$9,000 per MWh
    • Normal: ~$50–$100 per MWh
    • Event probability cited: ~30%
  • AI/power economics examples:
    • lab price point: $150/MWh (suggested too high)
    • plausible range: ~$50/MWh
    • plant capex: ~$80m–$100m (small plant; ~50 MW implied)
    • interest: ~20% YoY
    • production cost: ~$50/MWh
    • planning horizon: 6–12 months

Methodologies / step-by-step frameworks explicitly described

  • Hub pricing model pipeline

    • Forecast demand using weather
    • Map demand to dispatchable generation stack
    • Estimate expected hub price
    • Compare estimate vs market price
    • If mispricing persists after sanity checks → consider trade

  • Congestion/basis modeling approach (high level)

    • Estimate generation/demand at two locations
    • Model capacity limits and expected imports/exports
    • Optimize/solve the entire grid
    • Trade regional price spreads based on congestion likelihood

Presenters / sources mentioned

  • Neil Somani / Neel Somani / “Neil Smani” (guest; ex-Citadel quant; traded power markets)
  • Host (unnamed interviewer; runs the podcast conversation)
  • Citadel (referenced as employer; commodities group; discretionary power trading context)
  • Two Sigma (referenced as an automated systematic hedge fund example)
  • Onyx Capital Group (promotional segment; oil derivatives market maker; event/internship/competition)

Original video