Combining indicators without fooling yourself
सूचकहरू मिलाएर प्रयोग: कन्फ्लुएन्स र ओभरफिटिङ
Five indicators all flashing "buy" feels like strong confirmation. Often it's one piece of information counted five times. Here is how to combine tools properly - and how professionals avoid the trap of a backtest that only works in the past.
Five "buy" signals, one piece of information
RSI says oversold. Stochastic says oversold. CCI says oversold. Williams %R says oversold. It feels like four confirmations. It's really one: all four are momentum oscillators built from the same recent closes, in slightly different ways. When one says "oversold", the others almost always agree - they can't help it.
Combine across families
If you use several indicators, take one from each family - they measure different things, so agreement means more:
- Trend - is the longer trend with you? (a moving average's slope)
- Momentum - is the push strong or fading? (RSI)
- Volatility - how big are the swings, where does the stop belong? (ATR, Bollinger squeeze)
- Volume / flow - who's behind the move? (volume, the floorsheet)
- And underneath them all, the business - the leg charts can't see (the investing course, lesson 12: the three-legged chair).
The real danger: overfitting
With enough indicators and settings, you can build a rule that would have made money on any past chart. RSI 14 didn't work? Try RSI 9. Add a MACD filter. Only on Mondays. Only above the 50-day... Each tweak makes the backtest look better - because you're fitting the noise of the past, not a real pattern. Such rules look brilliant in testing and fail on new data. This is called overfitting, and it's how most "95% win rate" strategies are born.
While writing our e-books, we tested Candle Range Theory on NEPSE by stacking ideas together - setups, filters (trend, a real sweep, sector confirmation, early-week timing, market trend) and exits - 720 combinations. We picked the best combinations using data before July 2024, then tested them once on the data after.
The best in-sample combinations looked profitable. Out of sample, every one of them lost money, and only about 13% of all combinations were positive - roughly what chance alone produces. We withdrew the CRT book because of it. That's what honest testing looks like - and why you should be suspicious of any backtest you can't see tested on fresh data.
How to test without fooling yourself
- Decide the rule first, in writing, before looking at results.
- Keep it simple - few rules, few settings, standard values (RSI 14, 20-day bands). Fewer knobs, less fitting.
- Split your data: build on one period, test once on a later period you didn't touch.
- Compare with the base rate and look at the median and both tails (lesson 1).
- Include costs and T+2 - about 1% a round trip on NEPSE kills most small edges.
- Expect less live than in the test - always.
See it live on MacroLens NP
Everything in this lesson, worked out on real NEPSE data for every listed company - inside the MacroLens NP desk.
Lesson summary
- Many indicators are the same measurement in different clothes - RSI, Stochastic and CCI are all momentum.
- Combine across families: one trend, one momentum, one volatility, plus volume and levels - and fundamentals underneath.
- Overfitting: tuning rules until they fit past data perfectly. They look brilliant in the test and fail on new data.
- Always test rules on data you didn't use to build them. A real NEPSE example: a stack of CRT rules won before 2024 and lost after.
Create a free account to keep reading
Lesson 1 is open to everyone. From here on, the course is still completely free - we just ask for an account so we can save where you stopped, keep your quiz scores and let you pick up on any device.
No card, no payment. 9 more lessons, 89 minutes in all.
Education only, not investment advice: nothing here tells you to buy or sell any share. Investing carries risk - you can lose money. Examples use made-up companies unless a real one is named.