Inventory Management Algorithm: How Do Market Makers Decide How Much Gold to Hoard?

Core InsightIn March 2020, logistics disruptions caused by the COVID-19 pandemic led to a rare widening of the spread between New York gold futures and London spot gold prices. Due to disruptions in the physical gold transport, refining, and delivery chain, the spread that would normally converge through cross-market arbitrage widened rapidly, exposing institutions holding related cross-market positions to higher hedging and inventory risks. This event reminded the market that a market maker is not a machine capable of unlimited order throughput. The prerequisite for continuously providing two-sided quotes is that the algorithm can, based on volatility, holding costs, and risk limits, calculate the target inventory level and allowable deviation boundary in real time, and when inventory deviates, guide the position back to a safe range by adjusting the quote center.

In late March 2020, the premium of New York gold futures over London spot gold once reached approximately $70 per ounce, a rare spread level in recent years. The underlying reason was that the pandemic disrupted the global gold transport chain—refining capacity was limited, air transport reduced, making the transport and specification conversion of physical gold between London and New York more difficult.

For market makers, this event pushed inventory management from a "back-office issue" to the forefront. When hedging and delivery chains are disrupted, risks that could normally be quickly transferred or offset may be forced to stay on the institution's books. For example, in the Hong Kong Exchanges and Clearing Limited (HKEX) USD Gold Futures (GDU), Delta Horizon Capital（德合资本） and seven other market makers simultaneously posted two-sided quotes on the order book. Every trade changes the net position, and each position change tests inventory capacity and risk tolerance. If a market maker merely mechanically quotes two-sided prices and passively accepts executions, the accumulation of one-sided order flow will quickly push the position to an unsustainable level. This is precisely the reason for the existence of inventory management algorithms.

I. A Market Maker's Inventory Is Not Gold Bars in a Warehouse

When many people hear "inventory", their first reaction is physical gold bars stacked in a warehouse. However, in gold futures market making, "inventory" primarily refers to net position – the net contract position accumulated by the market maker through buying and selling that has not yet been offset by reverse trades or hedging.

Assume a market maker posts both bid and ask prices on the order book. When investors repeatedly click the ask price to buy, the market maker becomes the seller in each trade, continuously increasing its short position. If the market direction suddenly reverses – for example, if gold prices start to rise rapidly – these accumulated short positions will face directional losses. Conversely, if the market maker continuously passively buys, accumulating a long position, it will also face losses when gold prices decline.

Therefore, market makers do not let net positions grow indefinitely. One lot of the GDU contract corresponds to 1 kilogram of gold with a purity of no less than 9999, with the notional value varying with international gold prices. The larger the position, the greater the profit or loss for each dollar move in the gold price. The core task of the inventory management algorithm is to monitor the size and direction of the net position in real time during trading and try to keep it within the risk boundaries preset by the strategy.

What does this mean for you? The main goal of a market-making strategy is usually not to take long-term directional bets on gold prices, but to capture spreads and related trading income through continuous quoting while controlling the directional risk generated after trades. Inventory management helps avoid the accidental accumulation of one-sided positions that would expose the market-making strategy to excessive directional risk.

II. What Determines Target Inventory and Position Boundaries

The inventory management algorithm needs to answer two questions simultaneously: what target level the net position should be maintained at currently, and how far the position can deviate from this target at most. This number is not decided arbitrarily, but is the result of real-time calculation based on a set of dynamic variables.

First is volatility. When market volatility increases, the magnitude and frequency of price jumps both increase, and the uncertainty facing the position rises accordingly. Other things being equal, a rise in volatility usually increases the risk per unit of position; the system may therefore lower the target inventory, tighten position limits, or increase the risk compensation in quotes.

Second is holding cost. Holding net long or net short positions requires capital and margin. Futures positions require margin, and the related margin requirements may change with market risk and settlement arrangements. The larger the position, the more capital is tied up, and the opportunity cost of that capital increases. The inventory management algorithm needs to find a balance between quoting revenue and holding costs.

Finally is risk limits. In addition to applicable rules set by exchanges and clearing houses, institutions typically set internal boundaries such as intraday net position, overnight position, single product exposure, and loss limits based on their own risk policies. These hard limits are incorporated into pre-order risk controls and continuous monitoring to limit new risks.

What does this mean for you? Volatility, holding costs, and risk limits together determine how much "ammunition" a market maker has for quoting. This is not a static formula, but a real-time decision that continuously solves among market movements, position changes, and risk constraints.

III. When Inventory Deviates from Target, How Does the Algorithm Guide the Position Back?

The most sophisticated part of inventory management is not knowing "how much the current position is", but actively guiding the position back to the target level.

Assume a market maker's short position is already close to the preset upper limit. Buyers are still active in the market, investors continue to click the ask price to buy, and the short position keeps increasing. At this point, the inventory management algorithm can adjust both bid and ask prices upward as a whole. A higher ask price reduces the attractiveness for investors to continue buying from the market maker; a higher bid price makes it easier to attract sell orders to hit the order book — the market maker buys at a higher price, gradually reducing the short inventory through buying trades.

Conversely, if the market maker's long position is too high, the algorithm can adjust both bid and ask prices downward as a whole. A lower bid price reduces the probability of the market maker continuing to passively buy; a lower ask price makes it easier to attract buy orders to hit the order book — the market maker gradually reduces the long position through selling trades.

Through this shift of the quote center, the algorithm changes the relative attractiveness of trades in different directions — this is not achieved by pausing quotes or rejecting orders, but by naturally directing the order flow through the price mechanism. The larger the deviation, the larger the adjustment to the quote center; the smaller the deviation, the quotes gradually return to symmetry. This forms a self-feedback loop: Inventory deviation → Quote center shift → Change in relative attractiveness of trade direction → Gradual return of inventory to target range.

It should be noted that the shift of the quote center does not guarantee that order flow will reverse, it only changes the probability of trades in different directions. When volatility rises, the model typically re-evaluates inventory penalty, quote width, order quantity, and hedging speed simultaneously, rather than mechanically increasing or decreasing a single parameter.

What does this mean for you? The bid and ask prices posted by a market maker on the order book are not a pure judgment of "where the gold price should be", but also incorporate the dynamic factor of "in which direction my current position needs to be adjusted." Understanding this clarifies why a market maker's bid and ask quotes are not always perfectly symmetrical.

IV. Inventory Management and Risk Hedging: Two Things That Must Be Done Simultaneously

The shift of the quote center can help slow down the speed of inventory accumulation, but when facing sustained one-sided market pressure, merely adjusting quotes is insufficient. At this point, the market maker needs to simultaneously engage in active hedging.

In a hypothetical scenario, if an institution has accumulated short risk in the GDU contract, it may consider using related gold futures, spot, or other appropriate instruments to establish positions in the opposite direction to reduce price movement risk. The specific hedging instruments and execution methods depend on the institution's strategy, market conditions, and compliance requirements; not every institution adopts the same hedging arrangements.

The inventory management algorithm needs to consider two layers of decisions simultaneously: first, whether the current net position is within the target range, and second, the cost and efficiency of available hedging instruments. Hedging is not free — cross-market, cross-timezone hedging involves transaction costs, FX risk, time zone differences, and liquidity disparities. The algorithm needs to dynamically weigh the trade-off: whether to balance between quote adjustment and waiting for reverse trades, or to pay hedging costs to immediately reduce risk.

For professional liquidity providers including Delta Horizon Capital（德合资本）, mature inventory management not only requires the system to accurately identify position changes, but also requires the institution to dynamically weigh between quote adjustment, waiting for reverse trades, and active hedging based on market conditions.

What does this mean for you? The continuous two-sided quotes you see on the order book are not a reflection that market makers have "too much money to spend." Behind the continuously updated two-sided quotes on the order book, there is often the ongoing operation of quote center adjustment, risk hedging, and limit management.

Conclusion: Inventory Management Is Essentially Risk Control

For liquidity providers in HKEX gold futures, inventory management is not an optional "actuarial optimization" but a core mechanism that determines whether a strategy can continue to operate. The 2020 New York-London gold market dislocation showed that when transport, delivery, and cross-market hedging chains come under stress, the importance of inventory and risk management rises significantly.

A mature inventory management algorithm accomplishes three things simultaneously: real-time monitoring of net positions, dynamically calculating the target inventory and allowable deviation boundary based on volatility, holding costs, and risk limits, and guiding the position back to the safe range through quote center shifting and active hedging when inventory deviates. These tasks are continuously updated and interlinked during the trading process. Any lag in any link may amplify a controllable inventory deviation into an unbearable directional risk during a sudden market change.

For investors, the behind-the-scenes inventory management of market makers does not appear on the trading terminal, but it affects how institutions adjust quotes, control size, and maintain quoting capability when positions change and market pressure rises. The final spread and depth in the market are the result of exchange arrangements, market volatility, order flow, and the collective actions of multiple participating institutions. Futures and derivatives carry leverage risk; relevant technical capabilities do not mean investors can ignore market risks.

[1] Reuters, "Gold supply fears push spot prices far below U.S. futures", March 2020.https://www.reuters.com/article/business/gold-supply-fears-push-spot-prices-far-below-us-futures-idUSKBN21B224/
[2] Hong Kong Exchanges and Clearing Limited, USD Gold Futures contract specifications and liquidity provider list.https://www.hkex.com.hk/Products/Listed-Derivatives/Commodities/USD-Gold?sc_lang=en
[3] Hong Kong Exchanges and Clearing Limited, Derivative market margin information.https://www.hkex.com.hk/Services/Clearing/Listed-Derivatives/Risk-Management/Margin/Margin-Information?sc_lang=en