Learn to avoid indicator overload

Figure — Learn to avoid indicator overload

The Mathematics of Redundancy

Let's derive why adding indicators can actually reduce decision quality.

Suppose each indicator provides a signal with some accuracy $p$ (probability of correct signal). If indicators are independent, combined accuracy improves. But most technical indicators are correlated—they derive from the same price/volume data.

Correlation coefficient between two indicators $I_1$ and $I_2$ :

\rho_{12} = \frac{\text{Cov}(I_1, I_2)}{\sigma_{I_1} \sigma_{I_2}}

WHY this matters: If $\rho_{12} \approx 1$ (highly correlated), the second indicator adds almost zero new information but doubles your cognitive load.

Formula Shared Information Between Correlated Indicators

The correct way to measure information overlap is mutual information, not a linear subtraction. For two jointly Gaussian indicators with Pearson correlation $\rho$ :

MI(\rho) = -\tfrac{1}{2}\ln(1 - \rho^2)

Derivation from first principles:

Entropy of a Gaussian: A Gaussian variable $X$ with variance $\sigma^2$ has differential entropy $h(X) = \tfrac{1}{2}\ln(2\pi e\,\sigma^2)$ .
Joint entropy: For a bivariate Gaussian, the joint entropy uses the covariance-matrix determinant $|\Sigma| = \sigma_1^2\sigma_2^2(1-\rho^2)$ .
Mutual information = $h(X) + h(Y) - h(X,Y)$ . Plugging in and simplifying gives $MI = -\tfrac{1}{2}\ln(1-\rho^2)$ .

KEY property: $MI \to 0$ as $\rho \to 0$ (independent, no overlap) and $MI \to \infty$ as $|\rho| \to 1$ (fully redundant). The overlap is logarithmic, not linear.

Example: For $\rho = 0.89$ , $MI = -\tfrac{1}{2}\ln(1-0.89^2) \approx 0.39$ nats — a substantial but finite overlap. The two indicators are strongly redundant but not literally identical.

The Cognitive Load Tax (empirical heuristic)

Decision time grows faster than linearly with the number of indicators. This is a rule-of-thumb model, not a derived law: reading $n$ indicators is $O(n)$ , but reconciling conflicts requires comparing pairs, of which there are $\binom{n}{2}\propto n^2$ . A convenient interpolation people use is:

T_{\text{decision}} \approx k \cdot n^{1.5}

where $n$ = number of indicators, $k$ = base processing time. Treat the exponent 1.5 as an illustrative middle ground between linear ( $n^1$ ) and pairwise-quadratic ( $n^2$ ) growth — it is heuristic, not measured by Miller's Law (Miller's "7±2" concerns working-memory capacity, not time scaling).

Example calculation:

3 indicators: $T = k \cdot 3^{1.5} \approx 5.2k$
18 indicators: $T = k \cdot 18^{1.5} \approx 76.4k$

6× more indicators → how much slower? Under $T\propto n^{1.5}$ , multiplying $n$ by 6 multiplies time by $6^{1.5} \approx 14.7$ . So going from 3 → 18 indicators is roughly 15× slower, not just 6× — and that's before the added risk of freezing entirely.

Common Indicator Redundancies

Group 1: Momentum Oscillators (ρ ≈ 0.85-0.95)

All measure "speed of price change":

RSI (Relative Strength Index)
Stochastic Oscillator
Williams %R
CCI (Commodity Channel Index)

WHY they're redundant: All normalize recent price change to a bounded scale. The formulas differ slightly, but on the same asset, they track almost identically.

Worked example Worked Example: RSI vs Stochastic Correlation

Asset: AAPL daily, 2023-01 to 2023-12-31 Calculate correlation:

Compute RSI(14) for each day: $\text{RSI} = 100 - \frac{100}{1 + RS}$ where $RS = \frac{\text{avg gain}}{\text{avg loss}}$
Compute Stochastic(14,3,3): $\%K = \frac{C - L_{14}}{H_{14} - L_{14}} \times 100$
Pearson correlation: $\rho = 0.89$ Interpretation (done carefully): $\rho = 0.89$ means about $\rho^2 = 0.89^2 \approx 0.79$ , i.e. ~79% of the variance in one is explained by the other — not "89% of information." The true information overlap is logarithmic: $MI = -\tfrac12\ln(1-0.89^2)\approx 0.39$ nats. Either way the conclusion holds: the second oscillator adds little independent signal for double the effort.

Group 2: Moving Average Variants (ρ ≈ 0.95-0.99)

SMA (Simple Moving Average)
EMA (Exponential Moving Average)
WMA (Weighted Moving Average)
DEMA, TEMA, etc.

WHY they're redundant: All smooth price data with different weighting schemes. EMA vs SMA might differ by 1-2%, but trend signals (crossovers, slope) are nearly identical.

Group 3: Volatility Measures (ρ ≈ 0.75-0.90)

Bollinger Bands
ATR (Average True Range)
Standard Deviation bands
Keltner Channels

The underlying idea: All measure dispersion of price. Bollinger uses standard deviation, ATR uses true range, Keltner uses ATR on an EMA—but they spike and contract together.

The Optimal Indicator Portfolio

Why This Works: Orthogonal Information

If indicators measure orthogonal (uncorrelated) dimensions, the correlation matrix is near-diagonal:

\rho_{ij} \approx \begin{cases} 1 & i = j \\ 0 & i \neq j \end{cases}

When $\rho_{ij}\approx 0$ for $i\neq j$ , the mutual information between any pair is $-\tfrac12\ln(1-0^2)=0$ — each indicator contributes genuinely new information, so 3 uncorrelated indicators are worth far more than 10 correlated ones.

Symptoms of Indicator Overload

Checklist of Overload Symptoms

Analysis paralysis: Spend >10 minutes analyzing a simple swing trade setup
Conflicting signals: Indicators contradict each other >40% of the time
Can't explain your edge: When asked "Why did you enter?", you list 7 indicators instead of a clear thesis
Missed moves: Price moves 5% while you're waiting for indicator #8 to confirm
Frequent rule changes: Keep adding/removing indicators seeking the "perfect combo"

Practical Reduction Protocol

Worked example Case Study: Reducing12 Indicators to 3

Starting portfolio (actual beginner trader):

Trend: SMA(20), EMA(50), EMA(200), VWAP
Momentum: RSI(14), Stochastic(14,3,3), MACD(12,26,9)
Volatility: Bollinger Bands, ATR
Volume: OBV, Volume bars, VWAP (duplicate)

Step 1: Identify redundancies

SMA(20), EMA(50), EMA(200), VWAP → all trend → keep EMA(50) (intermediate timeframe)
RSI, Stochastic, MACD → all momentum → keep RSI(14) (simplest, bounded 0-100)
Bollinger, ATR → both volatility → keep ATR (useful for stop-loss sizing)
OBV, Volume, VWAP → keep OBV (cumulative, shows smart money flow)

Final portfolio: EMA(50), RSI(14), OBV

Result: Decision time dropped from avg 8min → 90 sec. Win rate improved (from 52% → 60%) because less noise → clearer patterns.

Why this step? EMA(50) catches the trend. Why RSI? Flags overbought/oversold extremes. Why OBV? Confirms if institutions are accumulating or distributing. Each answers a distinct question.

The Correlation Screening Test

Before adding an indicator, compute its correlation with your existing set.

Formula Correlation Screening Heuristic (rule of thumb, not a law)

For candidate indicator $I_{\text{new}}$ and existing set $\{I_1, \ldots, I_n\}$ :

\bar{\rho}_{\text{new}} = \frac{1}{n} \sum_{i=1}^{n} |\rho(I_{\text{new}}, I_i)|

Practical guide (these thresholds are heuristics chosen for convenience, not derived from theory — always backtest):

$\bar{\rho}_{\text{new}} < 0.5$ → likely adds new information
$0.5 \leq \bar{\rho}_{\text{new}} < 0.7$ → marginal value, test carefully
$\bar{\rho}_{\text{new}} \geq 0.7$ → probably redundant, likely reject

Why these numbers? They roughly map to shared variance $\rho^2$ : at $\rho=0.7$ , about half the variance ( $0.49$ ) is shared, which many traders treat as "too much overlap." Nothing magic about 0.7 — validate on your own data.

Example: You have RSI(14). Considering adding Stochastic.

Compute $\rho(\text{RSI}, \text{Stochastic})$ on 252 days of data → $\rho = 0.87$
Since $0.87 > 0.7$ (shared variance $\approx 0.76$ ) → likely reject Stochastic, it's redundant.

Advanced: Treating Indicators Like a Portfolio

Treat indicators like investments. Return = decision quality improvement. Cost = added complexity/confusion.

\text{Indicator value} \approx \frac{E[\Delta \text{Win Rate}]}{\text{added cognitive load}}

Heuristic: An indicator should improve your net results after accounting for the extra decisions you'll miss or delay due to added complexity.

Recall Explain to a 12-year-old

Imagine you're trying to decide if it's going to rain. You have:

A thermometer
A barometer
A hygrometer (humidity)
A weather app
Looking at clouds
Asking 5 neighbors what they think
Checking if your knee hurts (old injury aches before rain)

That's 7 "indicators"! But here's the problem: your neighbors are all looking at the same weather app, so asking 5 neighbors is really just checking the app 5 times. The thermometer and barometer are closely linked. Your knee and the clouds are the only different clues.

In trading: You want to know if a stock will go up. You can check if it's above its average price (trend), if it's moving too fast (momentum), and if big investors are buying (volume). That's 3 different clues. But if you check 5 different "momentum" indicators, you're just asking the same question 5 times. You feel busier, but you're not smarter—and now you're so busy you miss the rain (or the stock move) because you're still analyzing!

Connections

3.4.1-RSI-Relative-Strength-Index – Core momentum indicator, understand this deeply before adding others
3.4.8-Bollinger-Bands – Volatility measure, combines trend (MA) + volatility (SD)
3.3.2-Moving-Average-Crossovers – Trend system, can be complete strategy with just 2 MAs
2.5-Trading-Psychology – Analysis paralysis is psychological, not technical
4.2-Position-Sizing – ATR is useful here (stop-loss sizing), not for entry signals
5.1-Backtesting-Basics – Test if added indicators actually improve results
3.4.15-Building-a-Trading-System – Next step: combine your 2-3 indicators into rules

#flashcards/stock-market

What is indicator overload?

Using too many technical indicators (typically 4+) such that cognitive load increases faster than decision quality, leading to analysis paralysis and missed opportunities.

How do you correctly measure information overlap between two correlated indicators?

Use mutual information for jointly Gaussian variables:

MI = -\tfrac12\ln(1-\rho^2)

. It is logarithmic (goes to infinity as

\rho\to 1

, to 0 as

\rho\to 0

), NOT a linear subtraction of

\rho

Why is "correlation 0.89 = 89% of information shared" wrong?

Because

\rho

measures linear association, not information.

\rho=0.89

means ~

\rho^2=79\%

of variance explained, and the information overlap is logarithmic (

MI\approx0.39

nats), not linearly 89%.

Why are RSI and Stochastic Oscillator redundant?

Both are momentum oscillators measuring speed of price change on a bounded scale. Correlation is typically

\rho > 0.85

, so most of the variance (and information) is shared.

What is the optimal number of indicators?

2-3 indicators from distinct categories (trend, momentum, volume), chosen to be as uncorrelated as possible.

What correlation range is used as a redundancy rule of thumb?

|\rho| \gtrsim 0.7

(shared variance

\rho^2\approx0.49

) is a common heuristic for "too much overlap" — but it's not a theoretical law; always backtest.

Name three distinct indicator categories

Trend (e.g., moving averages), Momentum (e.g., RSI), Volume/Volatility (e.g., OBV, ATR)—each measures a different market dimension.

T\propto n^{1.5}

a proven law?

No. It's an illustrative heuristic interpolating between linear reading cost (

n

) and pairwise-comparison cost (

n^2

). Miller's Law concerns working-memory capacity (7±2), not time scaling.

Under

T\propto n^{1.5}

, how much slower is 6× more indicators?

6^{1.5}\approx 14.7

×, i.e. roughly 15× slower — not 6× and not 30×.

Why can more indicators decrease total wins?

Analysis paralysis from complexity causes you to miss trades. If added indicators reduce trade frequency by 40% but only improve win rate by 7%, net wins decrease.

What are three symptoms of indicator overload?

(1) Analysis paralysis (>10 min per setup), (2) conflicting signals (>40% of time), (3) missed moves while waiting for confirmation from too many sources.

Concept Map

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho yaar, is note ka core intuition bahut simple hai—jaise ek car chalate waqt agar tum ek saath 20 alag-alag gauges pe dhyaan doge, toh confuse ho jaoge aur accident ho jayega. Trading mein bhi bilkul same cheez hoti hai jise hum "indicator overload" kehte hain. Log sochte hain ki jitne zyada indicators use karenge utna safe rahenge, but reality ulti hai—har extra indicator noise add karta hai aur tumhari decision-making slow, confusing aur weak ho jaati hai. Isliye smart traders sirf 2-3 well-chosen tools rakhte hain jo alag-alag purpose serve karte hain.

Ab yeh kyun hota hai iska ek maths side bhi hai. Zyada tar technical indicators ek hi price aur volume data se bante hain, matlab woh aapas mein highly correlated hote hain. Agar do indicators ka correlation ( $\rho$ ) 1 ke paas hai, toh doosra indicator koi nayi information nahi de raha—bas tumhara mental load double kar raha hai. Isko measure karne ka proper tareeka mutual information hai, aur formula batata hai ki jaise-jaise correlation badhta hai, dono indicators ka overlap logarithmically badhta jaata hai. Simple language mein—do similar indicators tumhe wahi baat do baar bol rahe hain, alag insight nahi de rahe.

Sabse important baat yeh hai ki decision lene ka time linearly nahi, balki uss se tezi se badhta hai, kyunki har naye indicator ke saath tumhe usko baaki sab se compare karna padta hai. Rule-of-thumb ke hisaab se 3 se 18 indicators pe jaane ka matlab hai lagbhag 15 guna slower decisions—6 guna nahi! Isliye yeh cheez matter karti hai: kam but sahi indicators tumhe fast, clear aur confident banate hain, jabki zyada indicators tumhe freeze kar dete hain aur opportunities miss ho jaati hain. Ruthless curation seekho—yahi asli skill hai.

The Mathematics of Redundancy

Information Theory Foundation

The Cognitive Load Tax (empirical heuristic)

Common Indicator Redundancies

Group 1: Momentum Oscillators (ρ ≈ 0.85-0.95)

Group 2: Moving Average Variants (ρ ≈ 0.95-0.99)

Group 3: Volatility Measures (ρ ≈ 0.75-0.90)

The Optimal Indicator Portfolio

Why This Works: Orthogonal Information

Symptoms of Indicator Overload

Checklist of Overload Symptoms

Practical Reduction Protocol

The Correlation Screening Test

Advanced: Treating Indicators Like a Portfolio

Connections

Concept Map

Hinglish (regional understanding)

Hinglish (regional understanding)

Test yourself — Indicators & Oscillators