Using Order Flow to Estimate True Price
Order flow is the sequence of trades and quote changes arriving in the market. Read properly, it is the highest-frequency evidence available about where fair value is heading.
Signing the flow
The first step is deciding whether each trade was buyer-initiated or seller-initiated. A trade at the ask means an aggressive buyer crossed the spread; a trade at the bid means an aggressive seller did. Trades in between require classification rules, but the principle holds: the aggressor's direction is the information.
Aggregate this and you get order flow imbalance:
Persistently positive OFI is one of the more reliable short-horizon predictors of price movement, and it is a standard input to a market maker's fair value.
The signals worth watching
Size. Large trades are more likely to be institutional and informed. A sequence of small trades in the same direction can also be one large order being worked, which is why participants track cumulative signed volume rather than individual prints.
Persistence. One buy is noise. Twenty buys in a row against no sells is information. The autocorrelation of signed flow is high in real markets, partly because large orders are split up.
Book reaction. If aggressive buying occurs and the offer immediately refills at the same price, sellers are comfortable and the move may not persist. If the book thins and the offer steps up, the pressure is real.
Your own fills. The most direct evidence you have. Being repeatedly lifted means the market disagrees with your price, and it is disagreeing by trading with you.
The flow you receive is evidence about your own fair value. Consistently trading on one side is not good fortune, it is the market telling you your estimate is wrong.
Informed versus uninformed
The central difficulty: not all flow carries information, and the two look similar.
Uninformed flow comes from hedgers, index rebalancing, retail, and anyone trading for reasons unrelated to short-term value. This flow is profitable to trade against, and its price impact is temporary, so it mean-reverts.
Informed flow comes from participants who know something. Trading against it is a loss, and its price impact is permanent.
Distinguishing them in real time is the hardest inference in market making. The usual clues are the venue (retail-heavy venues carry less informed flow), timing (flow immediately before a scheduled announcement is suspicious), and the aftermath: if price reverts after a trade, the flow was probably uninformed; if it continues, it was not.
The mechanical caution
Some observed price movement is not information at all. The bid-ask bounce makes prices oscillate between bid and ask as buy and sell orders alternate, with no change in fair value whatsoever. A naive volatility estimate from trade prices overstates true volatility for exactly this reason.
Do not update fair value on a sequence that is simply alternating aggressors. Update on genuine imbalance.
Before treating a move as information, ask whether it is just the bounce. Alternating trades at bid then ask then bid is noise, however volatile the price series looks.
The loop
In practice this is Bayesian updating executed continuously: hold a fair value, observe flow, weight it by how informative you think it is, and revise. Update too slowly and you are picked off; update too fast and you chase noise into a worse price. Calibrating that speed is where experience shows.