Why is VaR or CVaR on high-frequency returns tricky?

Question

AcadiFi · Accepted Answer

High-frequency data creates more observations, but not automatically better risk estimates. At very short horizons, returns can be distorted by bid-ask bounce, minimum tick sizes, stale quotes, fees, latency, and changing liquidity. The distribution may look discrete, fat-tailed, and dependent in ways that standard daily-return models do not handle well.

Suppose Harborline Intraday Strategies estimates one-minute VaR for a market-making strategy. If the model treats every one-minute return as clean and independent, it may miss the effect of spread widening, quote flickering, or inventory constraints during stress. The historical sample may also include many zero or tiny moves followed by occasional jumps.

The FRM takeaway is that more data is not the same as more signal. High-frequency VaR requires careful cleaning, horizon choice, liquidity assumptions, and validation against actual execution. CVaR is especially sensitive because it averages the worst observations, which may reflect either genuine risk or data artifacts. Before trusting the number, ask whether the data represents tradable outcomes or just noisy quoted prices. On the exam, also state the horizon, confidence level, data source, modeling assumption, and validation step, because those choices usually explain why two reasonable VaR answers differ materially. Practice more in our FRM question bank.

Why is VaR or CVaR on high-frequency returns tricky?

Master Part II with our FRM Course

Related Questions

Practice Questions