The Reality of Living On The Bid-Ask Spread
Most people think market making is about quoting prices and collecting the spread. It's not. It's about managing inventory risk while getting run over by directional flow. The spread is just how you get paid for taking that risk. If you can't model your exposure quickly, you're not a market maker, you're a liquidity donor. In practice, this means running a continuous loop: quote two-sided prices, monitor your aggregate delta and gamma across every position, adjust when the underlying moves or volatility shifts, and decide when to hedge versus when to absorb. The risk analysis part is what actually keeps you alive. Quotes are easy. Knowing when you're about to blow up is the hard part. I'll walk through the mechanics, then get into the stuff that isn't in any textbook because nobody writes about failures publicly.
Setting Up The Quoting Engine
You need a system that takes real-time input from the underlying and options chains and outputs bid and ask prices. The pricing side uses Black-Scholes or a more accurate model like Heston if you're dealing with commodity options where skew is extreme. Greeks are computed at the same time. Your bid price is the model price minus your risk charge. Your ask price is the model price plus the risk charge plus the spread you want to earn. The risk charge isn't arbitrary. It's calculated from your inventory turnover, your volatility environment, and your hedge execution costs. During calm markets it might be 0.5 to 2 percent of the option premium. During earnings season or commodity supply shocks it can be 15 to 30 percent. If your risk charge doesn't widen when conditions deteriorate, you'll be the one buying the broken options at the wrong time. Latency matters less than you'd think. Sub-millisecond isn't necessary unless you're competing against HFT firms on exchange-provided liquidity. Real-time means under 200 milliseconds for most independent market makers. Beyond that, you're quoting stale prices and getting picked apart by arbitrageurs. I've seen shops lose three times their expected daily profit in a single session because their refresh cycle was set to one second instead of fifty milliseconds. The fix was trivial. The lesson wasn't.
Understanding Your Inventory Risk
Your primary exposure comes from delta and gamma. Delta tells you how much your position value changes per point move in the underlying. Gamma tells you how fast that delta changes. If you're short gamma, you're selling options and buying the underlying as it drops and selling as it rises. This feels like you're doing the right thing, but you're actually buying high and selling low relative to your average cost. That's negative convexity and it's where most retail market makers get destroyed. Commodity options add another layer because the underlying itself can gap. Natural gas futures have gapped 40 percent in a single session during weather events. Equity options gap less frequently but earnings moves can still produce 15 to 25 percent jumps. Your risk model needs to account for jump risk, not just diffusion risk. Standard Black-Scholes delta hedging assumes continuous price movement. It doesn't happen in commodities. A practical approach is to compute your portfolio VaR under historical scenarios and stress it with simulated jumps. Run it hourly at minimum. During high-volatility periods, every fifteen minutes. This gives you a number you can act on instead of a feeling that something might be wrong.
Get the Full Details
The Hedging Workflow
Delta hedging is the baseline. You hedge your net delta by trading the underlying or futures. The hedge ratio isn't always one-to-one. Commodity options often have basis risk between the option settlement price and the futures you're using to hedge. Soybean meal options settle against the CBOT soybean meal futures, but if you're hedging with crude-linked energy credits, your basis drift can eat your edge over time. Vega hedging is where most people quit. When implied volatility moves against your position, you need to buy or sell options to neutralize. The problem is that options are expensive to trade and illiquid in the face you need to trade them. During the March 2020 crash, vega hedging demand overwhelmed available liquidity. Bid-ask spreads blew out to 50 percent or more. The firms that survived had pre-arranged credit lines with counterparties who could take the other side when the public exchanges were useless. My rule of thumb: maintain a vega hedge ratio between 0.6 and 0.9 depending on liquidity conditions. Don't try to go to exactly zero vega. The transaction costs of chasing it will kill you. Accept a small residual vega exposure and manage it through position sizing rather than constant rebalancing.
A Specific Problem I Had With Commodity Option Skew
I was market making soybean oil option spreads during a period when USDA reports were causing massive skews. The puts were pricing with implied volatilities 30 to 40 percent higher than the calls at similar moneyness. My risk system was using a single volatility surface that didn't capture the skew properly. I was systematically underpricing the puts and overpricing the calls, which meant I was taking the wrong side of the market consistently. The fix was building a local volatility surface from the actual traded quotes rather than relying on the vendor-supplied pricing model. I spent two days pulling the full term structure across strikes and recalibrating. After that, my bid-ask alignment improved and my win rate on inventory flips went from roughly 48 percent to about 56 percent. That sounds small. Over a month of trading, it meant the difference between breaking even and making a 4 percent return on capital. The deeper issue was that the vendor model was smoothing the skew too much. It treated the market as if it wanted a single volatility curve. The market actually wanted two: one for the oilseed complex and one for the energy substitute relationship. Once I separated those regimes in the model, everything else fell into place.
Counter-Intuitive Things Beginners Miss
First, wider spreads don't always mean more profit. If your spreads are too wide, you get fewer quotes filled and your inventory builds up. Unfilled quotes are free money you never see. The optimal spread is the narrowest spread that still compensates you for your risk. Usually 1.5 to 3 times your model-implied transaction cost. Anything wider and you're leaving money on the table. Anything narrower and you're donating it. Second, gamma scalping works best when you're long gamma, not short. Short gamma traders think they can make money by collecting premium and hedging delta. They're not wrong in low-volatility environments. But when vol spikes, their gamma risk becomes exponential and their hedging costs exceed the premium they collected. Long gamma strategies lose to time decay slowly but win big during vol explosions. The asymmetry favors the long gamma side over multiple sessions. Third, your greatest risk isn't the market moving against you. It's the market staying still while your model drifts. When underlying volatility collapses and stays collapsed, your Greeks become inaccurate because they were calibrated to a higher vol regime. You're hedging based on wrong assumptions and your errors compound silently. I've seen this take down smaller market making operations because nobody noticed until the P&L statement showed it.
Practical Tools And Setup
You need real-time data feeds, a pricing engine, a risk calculation module, and an order management system. Data feeds can come from vendors like IQFeed, Polygon, or exchange-direct feeds. Pricing engines are typically custom-built in Python or C++. Risk modules need to handle multi-leg positions and compute sensitivities across time, price, and vol. Order management should support both manual and automated quoting with kill switches. For commodity options specifically, you need access to the underlying futures chain and a reliable way to map options to their delivery months. There are API wrappers available for most major exchanges. CME Group, ICE, and DTCC all provide programmatic access. The pricing libraries are less standardized. QuantLib works for equity options but struggles with commodity-specific features like storage costs, convenience yield, and seasonal patterns. I built a custom pricing module that handles commodity options by separating the forward curve construction from the volatility surface. The forward curve comes from the futures strip with roll adjustments. The vol surface is built from observed option prices using a regularized optimization that prevents negative variance. This took about three weeks to get right. Off-the-shelf solutions won't do this for you.
Where Market Making Breaks Down
This approach fails when you don't have sufficient capital to absorb inventory shocks. A single bad day in commodity options can require several hundred thousand dollars in hedging capacity. If your account is undercapitalized, you'll be forced to flatten positions at unfavorable prices regardless of what the model says. It also breaks down during exchange outages or data delays. I've been quoted and unable to cancel for forty-five seconds during a network hiccup. In that time, a silver futures spike moved the market 3 percent. When I got back online, I was flat on the wrong side. The system should have circuit-breaker logic that auto-flattens when data stops flowing. Most market makers don't implement this. They should. Regulatory constraints also limit effectiveness. Pattern day trader rules, margin requirements, and position limits vary by market and change frequently. Commodity options have different margin structures than equity options. Understanding these before you start is essential. Trading without this knowledge is how people get margin calls on positions that should have been fine.
Getting Started
Start paper trading. Run your model against live data without committing capital for at least two weeks. Track your quoted spreads, fill rates, inventory buildup, and theoretical versus actual P&L. The gap between theoretical and actual P&L is your real cost of doing business. It includes slippage, hedge execution quality, and model error. If that gap is larger than your expected profit per trade, your model needs work before you go live. When you're ready to go live, start with one product. One underlying. One options chain. Get comfortable with the inventory management and hedging workflow before adding complexity. Adding a second product while you're still learning the first one is how people lose money fast. The risk analysis framework stays the same regardless of product. Delta, gamma, vega, theta, and inventory VaR are the core metrics. Everything else is optimization. Get the core right and the margins will follow. Get them wrong and no amount of optimization will save you.
