The relative volatility index as a filter

[Fairman]

The relative volatility index as a filter

The relative volatility index looks like RSI and is built in a similar way, but it uses the standard deviation of price instead of price changes. It measures the direction of volatility. Is volatility expanding on up moves or on down moves?

A reading above 50 means volatility has been higher on up bars than on down bars over the lookback. Below 50, the reverse.

On its own it's not a signal. Its original use, and still the more practical one, is as a confirmation filter for another indicator or for price action. For example, only taking long setups when the RVI is above 50, and short setups when it's below. That keeps you trading in the direction where the bigger moves have been happening.

On M5 it reacts quite smoothly, which is a reason some traders prefer it over RSI as a filter. It's less jumpy on individual bars and less likely to flip back and forth during a choppy period.

The trap is reading it like an overbought or oversold indicator. A high RVI doesn't mean price will turn. It means up moves have been more volatile, which in a trend is exactly what you'd expect. Fading a high reading in a trend is fighting the market.

[FXS London Scalper]

Using the RVI as a direction filter rather than a trigger is the sensible way to apply it. It also helps around news: at 13:30 London a single large US data bar can flip the RVI across 50 on M5, and for the next 14 bars the filter mostly reflects that one candle. Ignoring the filter for a short window after high impact releases, or using a longer lookback on those days, keeps one spike from deciding the direction for the rest of the afternoon.

[Fairman]

The lookback for the standard deviation and the smoothing period both affect the reading. Common defaults use a 10 period standard deviation and a 14 period smoothing. Shortening either makes it react faster, but also brings more noise. For a scalper trading M1 or M5 entries but using M15 for direction, putting the RVI on M15 with default settings is a sensible starting point. It gives a direction read that's stable over the time frame of a typical scalp, without flipping every few bars on the entry chart where you'd otherwise be distracted by it. Whatever settings you use, keep them the same across pairs so readings are comparable.

[Fairman]

Divergence between RVI and price can be informative, though it's not as reliable as people often claim. If price makes a new high but the RVI makes a lower high, volatility on the latest up move was smaller than on the previous one. That's a sign the push is weaker. It doesn't mean the trend is over. In strong trends, divergences can stack up for a long time without any reversal. Treating divergence as a reason to tighten stops or take partial profit, rather than as a reason to reverse, is usually a more realistic way to use it. Writing down each divergence you notice and what price did over the next hour is a quick way to check how much it means on your pairs.

[Fairman]

Comparing RVI readings across correlated pairs can be useful. If EURUSD and GBPUSD both show RVI above 50, the dollar weakness is broad. If one is above and the other below, the move is more specific to one of the other currencies. That can help decide which pair to trade. The pair where the RVI agrees with your planned direction is the one where volatility is supporting you, which tends to give cleaner follow through after entry and fewer moments where the trade stalls just past your entry price. Gold against the dollar pairs can work the same way, since gold often moves against the dollar and its RVI can confirm or contradict the dollar read.

[Fairman]

Some traders use the RVI alongside RSI as a pair. RSI measures the direction of price change. RVI measures the direction of volatility. When both agree, say both above 50, the move has both momentum and expanding volatility behind it. When they disagree, the move is less convincing. It's a simple check and doesn't add much clutter, since both indicators can share one subwindow on the same 0 to 100 scale, which keeps the chart readable while adding a second point of view on the same move. It's also a good way to avoid adding yet another indicator window, which is a real concern on a crowded scalping layout.

[FXS Prop Desk]

Good explanation of the RVI. As a filter, its most practical use is probably in agreement with the trade direction: taking longs only when volatility is expanding on up bars. The tricky part is choosing the lookback. A short lookback on M5 makes it react to every few candles, which is close to noise, while a long one turns it into a description of the day so far. Testing two or three values and keeping the one that leaves enough trades is safer than optimising it closely.

[Shadow Trader]

A scalper tests a simple entry filter: only take longs when, over the last 50 ticks, the ask moved up more often than the bid moved down. Backtested on a couple of weeks of recorded ticks, it trims the losing trades noticeably. Win rate rises, drawdowns shrink, and the equity curve smooths out.

Then it goes live and the benefit fades. The twist turned out to be in the data, not the idea. The recorded ticks came from the platform's own history, where consecutive quotes with the same price had been collapsed. Live, the platform delivered every update. The count of upticks over a fixed number of ticks meant something else, because the window now covered far less time on the live feed than in the test. The filter was effectively measuring a shorter and noisier slice.

Fixing it meant changing the window from a tick count to a time window, say the last ten seconds, and recording ticks live rather than relying on exported history. After that the live results lined up with the test much more closely, though the edge was smaller than the original curve suggested.

The takeaway is that tick-based filters are sensitive to how the feed is built, so the same code can mean different things on different brokers or even different account types at one broker.

Has a tick-level filter ever behaved differently for you between backtest data and live, and what turned out to be the cause?