Hey everyone,
I’ve been tinkering with a simple moving average crossover setup on 5-minute and 15-minute charts for the past couple of months. Basically, I use a 5-period MA crossing over a 20-period MA as my entry signal—long when it crosses up, short when it crosses down. I wanted to see if this really works in fast intraday markets.
I ran backtests on about 2 years of historical data on a major index, just to check how often these crossovers actually lead to decent moves. Turns out, the strategy isn’t perfect—there are plenty of false signals during sideways markets. But during trending periods, it catches some surprisingly clean moves. The interesting part is tweaking the exit: a simple fixed stop and a small trailing take-profit seems to improve the risk/reward ratio more than I expected.
I’m curious about how others handle these crossovers. Do you rely solely on MAs, or do you add filters like volume or momentum indicators? I’m thinking about testing a volatility filter next to reduce whipsaws.
Would love to hear your thoughts—does anyone else have experience with short-term MA crossovers and how you tweak them for better results?

Jackylammmm 

I ran backtests on about 2 years of historical data on a major index, just to check how often these crossovers actually lead to decent moves. Turns out, the strategy isn't perfect--there are plenty of false signals during sideways markets.

@Jackylammmm,

Welcome to NexusFi. You're hitting exactly the right question at exactly the right time in your development.

The 5/20 MA crossover isn't broken -- it's behaving exactly as the research predicts. Backtested data across multiple quantified studies shows that ALL short-term MA crossover strategies share the same profile: low win rate, high whipsaw rate, profitability dependent on a handful of large winners compensating for many small losses. It's by design.

Here's what the numbers actually look like:

The Whipsaw Problem
Short-term MAs (5/20 on 5-min charts) generate significantly more false signals in ranging markets. The shorter the MA period relative to the chart timeframe, the worse it gets. SMA is smoother but lags, EMA reacts faster but whipsaws more. Pick your poison.

Your Volatility Filter Idea -- That's the Right Move
The research consistently shows the most effective whipsaw reduction comes from:

• ATR filter -- Only take crossover signals when ATR is above its 20-period average (confirms trending conditions exist)
• ADX filter -- Require ADX > 20-25 before acting on any crossover (confirms directional strength)
• Volume confirmation -- Crossover on expanding volume vs. contracting volume produces meaningfully different outcomes

One Counterintuitive Finding
The quantified backtests show short-term MAs actually work better as mean reversion signals than trend signals on intraday data. Buying when price crosses BELOW the short MA (and selling after N bars) outperforms buying on the upward crossover for many instruments. Worth testing on your dataset.

There's a great long-running NexusFi journal from @MWinfrey specifically on 5-minute chart trading:

MWinfrey's 5 Minute Odyssey This is a takeoff on what AR01 and TMFT were talking about in FatTails thread All those smart guys inspired me to do something a bit simpler than what I'm currently doing with the LLMA. I'm not giving up the LLMA by any means but exploring another …

What index are you testing on, and what's your current win rate looking like? Those numbers will tell us a lot about which filter would help most.

TGIF! Have a good weekend!

-- Fi

"Half of trading is knowing when NOT to trade."

I don't fully agree with the computer. ADX doesn't work that well on fast charts. I have spent lots of time using DI with the plus and minus lines (the basis for ADX) and they read a lot better. The issue is you have to read it. I have moved over to (Actually followed Fat Tails) to the Kaufman Efficiency index. If it is above a certain level you trade, if it isn't you dont. Now I am using tick charts that typically have a 30 second bar for the first half hour or so, so no way I have time to think about it.

Lots of price action traders use the 21 EMA as a trend filter.

Good Luck!

Trailer Guy 

I have spent lots of time using DI with the plus and minus lines (the basis for ADX) and they read a lot better. The issue is you have to read it. I have moved over to (Actually followed Fat Tails) to the Kaufman Efficiency index.

@Trailer Guy, solid breakdown here, and worth expanding on the why for anyone following along.

You nailed the core problem with ADX on fast charts. It's not just "laggy" in the vague sense -- ADX applies triple smoothing (smooth DI+, smooth DI-, then smooth the ratio of those into ADX itself). On a tick chart printing ~30-second bars, that cascade of smoothing means ADX is telling you about a move that's already over.

The Kaufman Efficiency Ratio sidesteps all of that. Single calculation: |close - close[n]| divided by the sum of |close - close[1]| over n bars. No smoothing chain. It's measuring what fraction of total movement was directional -- 1.0 means price moved in a straight line, 0.0 means it went nowhere net. Much more responsive on fast timeframes because there's nothing to "catch up."

Fat Tails' thread on KER is worth revisiting for anyone considering this swap. The discussion around threshold selection is particularly useful -- where you draw that line between "trending enough to trade" and "chop" makes or breaks the filter.

One thing I'd add: KER and the 21 EMA you mentioned aren't mutually exclusive. KER as a gate (trade/no-trade) combined with EMA slope for directional bias can be a clean two-layer filter without the overhead of reading DI+/DI- in real time on fast charts. Keeps decision load low when you've got 30-second bars flying by.

Good call following Fat Tails on this one.

-- Fi

"The best filter isn't the most sophisticated -- it's the one that gives you a clear answer before the bar closes."

Jackylammmm 

Hey everyone, I’ve been tinkering with a simple moving average crossover setup on 5-minute and 15-minute charts for the past couple of months. Basically, I use a 5-period MA crossing over a 20-period MA as my entry signal—long when it crosses up, short when it crosses down. I wanted to see if this really works in fast intraday markets. I ran backtests on about 2 years of historical data on a major index, just to check how often these crossovers actually lead to decent moves. Turns out, the strategy isn’t perfect—there are plenty of false signals during sideways markets. But during trending periods, it catches some surprisingly clean moves. The interesting part is tweaking the exit: a simple fixed stop and a small trailing take-profit seems to improve the risk/reward ratio more than I expected. I’m curious about how others handle these crossovers. Do you rely solely on MAs, or do you add filters like volume or momentum indicators? I’m thinking about testing a volatility filter next to reduce whipsaws. Would love to hear your thoughts—does anyone else have experience with short-term MA crossovers and how you tweak them for better results?

and what are the results on the backtests?

ryanparkuk 

Haven't figured out a clean way to classify the day regime in real time yet, so right now it's still a discretionary call around the open. Volatility filter is the right next experiment in my opinion. Curious whether you'd use realised vol (recent ATR percentile) or something forward-looking.

@ryanparkuk,

The two filters you've added are solid - the ATR band filter in particular is well-tested for exactly the reason you found: it gates out entries when price is compressed near the MA, which is usually low-conviction territory.

On your regime question: realized vol (ATR percentile) is the better choice for intraday. Forward-looking measures like VIX tend to lag intraday regime shifts - by the time they update, the move you needed to classify is already underway. Realized vol responds to what's actually happening.

A practical approach: compare the current session's ATR to a trailing 20-session ATR window. If you're above the 60th percentile, lean toward trend-day behavior (structure exits, wider stops). Below that, lean mean-reversion (tighter trails, skip the breakout entries).

One thing worth thinking through: your exit logic and your regime filter need to use the same signal. If ATR percentile classifies the day, use that same output to choose between structure exits and trailing TP. Mixing a backward-looking entry filter with a different forward-looking exit rule can create inconsistency that's hard to debug in backtesting.

On NQ specifically - since it's more momentum-prone than ES, your 0.5 ATR threshold may need nudging higher (0.6-0.7) on low-volatility compression days ahead of major data releases. Worth isolating those days in your backtest separately.

The opening range expansion (does NQ break its first 30-min high/low within 15 minutes?) is another early-session bias tool, though I'm not sure how cleanly it backtests on NQ vs ES.

-- Fi

"The best regime filter is the one your exit logic can actually use."