Order Flow Imbalance (OFI)

The foundational data structure from which OFI is derived. The LOB is an electronic record of all outstanding buy and sell orders for a specific asset, organized by price level. OFI is calculated by observing the aggressive trades that consume liquidity from the LOB's best bid and ask levels over a specified interval.

• Bid Side: Aggressive selling hits the bid.
• Ask Side: Aggressive buying lifts the offer.
• Depth: OFI can be extended to multiple levels of the book.

An alternative data sampling technique that creates bars not by fixed time or volume, but when the amount of new information reaches a predefined threshold. OFI is a primary metric used to trigger these bars, as a high absolute imbalance signals the arrival of informed, directional order flow.

• Tick Imbalance Bars: Sample when the cumulative signed tick imbalance exceeds a threshold.
• Volume/Dollar Bars: Sample based on cumulative volume or dollar value, often combined with OFI logic.
• Benefit: Produces more samples during high-activity periods and fewer during low-activity lulls.

A metric that estimates the fraction of informed trading in a market by analyzing persistent imbalances in volume-synchronized buckets. While OFI measures the net aggressive volume, VPIN uses the distribution of that volume across time to infer the probability of toxic order flow, often serving as an early warning signal for liquidity crises.

• Volume Buckets: A fixed amount of volume defines each observation.
• Toxicity: High VPIN indicates a high probability of adverse selection for market makers.
• Flash Crash Prediction: VPIN spiked dramatically before the 2010 Flash Crash.

A self-exciting point process where the occurrence of an event increases the probability of future events in the near term. OFI events are not independent; a large buy imbalance often triggers further buying. Hawkes processes model this clustering behavior to forecast the intensity of future order flow.

• Kernel Function: Defines how the influence of a past event decays over time (often exponential).
• Mutually Exciting: A multivariate Hawkes model can capture how buy OFI excites sell OFI and vice versa.
• High-Frequency Application: Used to model the arrival of trades and quote updates at the millisecond level.

The adverse effect on the price of an asset caused by the act of trading it. OFI is a direct driver of temporary market impact. A sustained buy imbalance signals aggressive demand, pushing the price up as liquidity providers adjust their quotes to avoid adverse selection. This impact has two components:

• Permanent Impact: The price change due to the information conveyed by the trade.
• Temporary Impact: The price concession needed to attract liquidity, which often reverts.
• OFI as a Predictor: Empirical studies show OFI is a strong, short-term predictor of price changes due to this impact mechanism.

A deep learning architecture that uses dilated, causal convolutions to model sequential data. TCNs are a popular choice for forecasting price movements from OFI sequences because they can capture long-range dependencies without the vanishing gradient issues of RNNs and offer parallel computation for fast training.

• Causal Padding: Ensures no information leakage from the future.
• Dilated Convolutions: Exponentially increase the receptive field to capture long-term patterns in OFI streams.
• Residual Connections: Stabilize training in very deep networks, allowing the model to learn complex non-linear relationships between OFI and future returns.