Handling Post-News Spikes in Digital-Filter Forex Signals

Summary

The question concerns a spot forex system that uses digital filters and behaves poorly around news releases. Rejecting signals generated within two hours of an announcement has not resolved the issue, because signals shortly afterward can still react to large price spikes. The response says the right treatment depends on how the filter is fitted, whether it is linear, and which nonlinear effects it accounts for.

The response frames post-news behavior as either non-stationarity within the same process or a distinct market process. The historical window used to estimate current conditions affects how long trading should remain disabled after an announcement. It also suggests adapting model parameters to an external variable associated with news. No specific filter design, test, or rule is provided, so the guidance is diagnostic: model assumptions and the duration of regime changes must be examined before choosing a spike-handling method.

Key ideas

• Post-announcement signals may remain distorted even when signals near the release are excluded.
• The suitable response depends on the filter’s fitting assumptions and its linear or nonlinear structure.
• News-driven moves can be modeled as non-stationarity or as a separate process.
• The estimation window helps determine how long a system should stay inactive after news.
• An external variable may be used to adapt filter parameters to news conditions.

Tags

# interpreting huge jumps


# interpreting huge jumps












i have been working on this trading system that uses digital filters to generate signals. the system works fine during normal market hours. but it goes haywire when there is news release. i have specifically ensured that all signals generated within 2 hours of a news release be rejected. the problem is how to interpret the signals coming a few minutes after a news release. i was just wondering if there is a way of treating the huge spikes caused due to news release. any insight would be helpful. PS: the system trades spot forex.

## Answer by htrahdis (score 0)

https://quant.stackexchange.com/a/9421

You have to mention more details as in what are your assumptions when you fit the digital filter. Also whether is it linear or not. If not then what non-linearities have you taken into account. The answer depends on that.

Because you can either look at it as a form of non-stationarity or a different process altogether. Also how much time behind are you looking at to model the current situation, That will decide how long should you switch off the system after news has come. Finally you can make your parameters adaptable to some external variable to try and adjust for the news.

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.