News Isn't The Problem

A news filter can make a strategy feel safer while cutting out the exact windows where it earns its edge.

Stick a news filter onto a strategy and you feel "safer" almost immediately. Block the high-impact releases, avoid the volatility. It sounds like obvious risk management. For a good number of strategies, it can remove the edge.

The dark side of a news filter

Here is the problem that I found, from pure ignorance for a long time. Some strategies make most of their money precisely because of how price behaves around scheduled events, or the volatility that then ensues. A breakout system on indices, a momentum model on comms, anything that feeds on the volatility expansion news creates, these earn their place during the exact windows whereby a news filter would tell them to stand down.

The reason this goes unnoticed is that the filter improves the metric everyone looks at first. Drawdowns look smoother. The large "news-day" candles vanish from the equity curve. It feels like you tightened the strategy up. What you actually did was lower the return more than you lowered the risk, so the Sharpe got worse even though the curve got prettier.

The diagnostic almost nobody runs

Per-session, per-event trade attribution. Tag every backtest trade with the session it opened in and its proximity to a scheduled high-impact release. Then split the P&L and answer one question: how much of this strategy's total profit comes from trades that opened in the news windows the filter would have blocked?

If the answer is barely any, fine, the filter is close to free and you can run it for the slippage protection. If the answer is most of it, the filter is not protecting the strategy. It is performing a lobotomy and charging you for the anaesthetic. The only way to know which case you are in is to look, because the two are indistinguishable from the equity curve alone.

The tail dependency problem

A strategy can be net-positive in news windows on the back of one or two enormous outlier trades, with the median news trade a loser. That strategy has a real tail dependency on events, and filtering it changes its entire character, average and tails alike. Attribution shows you the distribution. The sign on its own hides it.

I run this split on anything with a session or event component before I trust a filter decision either way. It takes an afternoon, the data is already sitting in your backtest, and it swaps a reflex, news is dangerous, filter it, for a measurement: here is what news is actually worth to this specific strategy.

A news filter is a real tool. Just don't install it blind. As with anything: data first.