Separating Volatility and Order Flow in Market-Making Models
Summary
The discussion explains why constant volatility appears in market-making models and how volatility differs from the risk that incoming trades carry adverse information. Volatility estimates the risk that an inventory position will lose value as prices move; trade arrival rates affect how quickly quotes are filled and whether fills worsen inventory imbalance. These risks call for different model inputs.
For predictable events, such as scheduled economic announcements, a volatility forecast can reflect anticipated price uncertainty and encourage the model to reduce inventory beforehand. Directional news calls for adjusting the estimated bid and ask fill probabilities, since informed flow may favor one side. The questioner reports poor backtests using Avellaneda–Stoikov models on EUR/USD tick data, but the response offers no tested adjustment or performance evidence. It gives modeling guidance rather than a complete calibration procedure, and it does not address data quality, execution costs, or other reasons a backtest may disappoint.
Key ideas
- Volatility models inventory exposure to future price moves, while fill probabilities model trading activity.
- Scheduled events can be represented as periods of elevated expected volatility.
- Directional information can make order flow asymmetric between bid and ask quotes.
- A volatility adjustment alone may not capture adverse selection from informed traders.
Tags
Full text
# Market Making constant volatility assumption # Market Making constant volatility assumption I have read a few papers on market making and all(nearly) assume that the stock follows a brownian motion with no drift and constant volatility.These assumptions seems un-intuitive to me because of intraday volatility and garch effects. I also downloaded some free tick by tick data of EURUSD from internet and ran the market making models of avellanda and stoikov on it and the backtest results were quite poor. Can any practitioner or anyone provide some tips or methods to adjust the volatility for my project ,any reference material would also be appreciated. ## Answer by lehalle (score 1) https://quant.stackexchange.com/a/73698 That's true that in market making papers, like Dealing with the Inventory Risk. A solution to the market making problem 2012 by Guéant, L and Fernandez-Tapia, the volatility is often taken as constant. What does it means? indeed you have two nature of uncertainties on the table when you make the market: - the mark to maket value of your inventory. from a quantitative viewpoint on market making, the volatility is the proxy that your inventories goes against you. The volatility is a way to mesure the expected value of your inventory in the future. - the probability that other market participants will trade with you, that leads to two effects: it allows to reduce your inventory (reacting to an anticipation of risk, thanks to your expectation of the volatility), or it could inflate your inventory towards more imbalance in an undesired solution (this is typically measure the adverse selection risk, that very often practitioners call toxicity of the flow). They are very different in nature. If you know something about future prices, for instance thanks to a calendar (some economic numbers will be disclosed during the trading day, at 14:30 for Europe, but you do not know if they will be positive or not): put this information in your volatility model. It will naturally tell to the optimizer that you have to reduce your inventory before 2.30pm. If you have a directional information on the price, for instance a News has just been disclosed about a patent issued by a company (think about the announce of Pfizer 's covid vaccine), it is affecting the flow of other participants in the direction of the information: if your quotes do not adjust, they will be hit a bad way. They you have to put this in your model of the probability of being lifted (that is often an point process with asymmetric intensities on the bid and ask side).
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.