Learn to avoid indicator overload

3.4.14 · Stock-Market › Indicators & Oscillators

Figure — Learn to avoid indicator overload

The Mathematics of Redundancy

Chalte hain kyun indicators add karne se decision quality actually kam ho sakti hai, yeh derive karte hain.

Maano ki har indicator ek signal deta hai kuch accuracy $p$ ke saath (correct signal ki probability). Agar indicators independent hain, toh combined accuracy improve hoti hai. Lekin zyaatar technical indicators correlated hote hain—woh sab usi price/volume data se derive hote hain.

Correlation coefficient do indicators $I_{1}$ aur $I_{2}$ ke beech:

ρ_{12} = \frac{Cov ( I _{1} , I _{2} )}{σ _{I_{1}} σ _{I_{2}}}

KYUN matter karta hai: Agar $ρ_{12} \approx 1$ (highly correlated), toh doosra indicator almost zero new information add karta hai lekin tumhara cognitive load double kar deta hai.

Formula Correlated Indicators ke Beech Shared Information

Information overlap measure karne ka sahi tarika mutual information hai, linear subtraction nahi. Do jointly Gaussian indicators ke liye jinka Pearson correlation $ρ$ hai:

M I (ρ) = - \frac{1}{2} ln (1 - ρ^{2})

First principles se Derivation:

Gaussian ki Entropy: Ek Gaussian variable $X$ jiska variance $σ^{2}$ hai, uski differential entropy $h (X) = \frac{1}{2} ln (2 π e σ^{2})$ hoti hai.
Joint entropy: Bivariate Gaussian ke liye, joint entropy covariance-matrix determinant $∣Σ∣ = σ_{1}^{2} σ_{2}^{2} (1 - ρ^{2})$ use karti hai.
Mutual information = $h (X) + h (Y) - h (X, Y)$ . Plug in karke simplify karne par $M I = - \frac{1}{2} ln (1 - ρ^{2})$ milta hai.

KEY property: $ρ \to 0$ par $M I \to 0$ (independent, koi overlap nahi) aur $∣ ρ ∣ \to 1$ par $M I \to \infty$ (fully redundant). Overlap logarithmic hai, linear nahi.

Example: $ρ = 0.89$ ke liye, $M I = - \frac{1}{2} ln (1 - 0.8 9^{2}) \approx 0.39$ nats — yeh substantial lekin finite overlap hai. Do indicators strongly redundant hain lekin literally identical nahi.

The Cognitive Load Tax (empirical heuristic)

Decision time indicators ki sankhya ke saath linearly se zyada tezi se badhti hai. Yeh ek rule-of-thumb model hai, derived law nahi: $n$ indicators padhna $O (n)$ hai, lekin conflicts reconcile karne ke liye pairs compare karne padte hain, jinki sankhya $(2 n) \propto n^{2}$ hai. Logon ka commonly use kiya jaane wala ek convenient interpolation hai:

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

jahan $n$ = indicators ki sankhya, $k$ = base processing time. Exponent 1.5 ko linear ( $n^{1}$ ) aur pairwise-quadratic ( $n^{2}$ ) growth ke beech ek illustrative middle ground samjho — yeh heuristic hai, Miller's Law se measure nahi kiya gaya (Miller ka "7±2" working-memory capacity ke baare mein hai, time scaling ke baare mein nahi).

Example calculation:

3 indicators: $T = k \cdot 3^{1.5} \approx 5.2 k$
18 indicators: $T = k \cdot 1 8^{1.5} \approx 76.4 k$

6× zyada indicators → kitna slow? $T \propto n^{1.5}$ ke under, $n$ ko 6 se multiply karne par time $6^{1.5} \approx 14.7$ se multiply ho jaata hai. Toh 3 → 18 indicators jaana roughly 15× slow hota hai, sirf 6× nahi — aur yeh pehle ki baat hai ki poori tarah freeze hone ka risk bhi hai.

Common Indicator Redundancies

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

Sab "price change ki speed" measure karte hain:

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

KYUN redundant hain: Sab recent price change ko ek bounded scale par normalize karte hain. Formulas thode alag hain, lekin same asset par yeh almost identically track karte hain.

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.

KYUN redundant hain: Sab price data ko alag-alag weighting schemes se smooth karte hain. EMA vs SMA mein 1-2% ka fark ho sakta hai, lekin trend signals (crossovers, slope) almost identical hote hain.

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

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

Underlying idea: Sab price ki dispersion measure karte hain. Bollinger standard deviation use karta hai, ATR true range use karta hai, Keltner ek EMA par ATR use karta hai—lekin yeh sab saath spike aur contract karte hain.

The Optimal Indicator Portfolio

KYUN Kaam Karta Hai: Orthogonal Information

Agar indicators orthogonal (uncorrelated) dimensions measure karte hain, toh correlation matrix near-diagonal hoti hai:

ρ_{ij} \approx {10 i = j i \neq = j

Jab $i \neq = j$ ke liye $ρ_{ij} \approx 0$ hota hai, kisi bhi pair ke beech mutual information $- \frac{1}{2} ln (1 - 0^{2}) = 0$ hoti hai — har indicator genuinely new information contribute karta hai, isliye 3 uncorrelated indicators 10 correlated walon se kahin zyada valuable hain.

Indicator Overload ke Symptoms

Common mistake Common Mistake: "Zyada Indicators = Zyada Confidence"

Galat logic: "Agar RSI buy keh raha hai AUR Stochastic buy keh raha hai AUR CCI buy keh raha hai, toh main 3× zyada certain hoon!"

KYUN sahi lagta hai: Multiple sources se confirmation psychologically comforting hota hai (consensus heuristic).

Reality: Agar teeno momentum oscillators hain ( $ρ > 0.85$ ), toh tumne effectively wahi sawaal teen baar poocha hai. Yeh 3 independent votes nahi hain; yeh ~1 vote hai jo teen baar count kiya gaya — shared information (mutual information se measure ki gayi) bahut zyada hai, isliye unka "independent confirmation" kaafi hadd tak illusion hai.

Fix: Steel-man the mistake: Tumhe confirmation chahiye—lekin uncorrelated sources se (trend + momentum + volume), na ki same calculation ke redundant variants se.

Overload Symptoms ki Checklist

Analysis paralysis: Ek simple swing trade setup analyze karne mein >10 minutes lagana
Conflicting signals: Indicators >40% time ek doosre se contradict karte hain
Apna edge explain nahi kar sakte: Jab pucha jaye "Tune entry kyun li?", 7 indicators list karo instead of ek clear thesis
Missed moves: Price 5% move kar jaata hai jab tum indicator #8 ke confirm karne ka wait kar rahe ho
Frequent rule changes: "Perfect combo" dhundne mein indicators add/remove karte rehte ho

Practical Reduction Protocol

Worked example Case Study: 12 Indicators se 3 par Aana

Starting portfolio (ek actual beginner trader ka):

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: Redundancies identify karo

SMA(20), EMA(50), EMA(200), VWAP → sab trend → EMA(50) rakho (intermediate timeframe)
RSI, Stochastic, MACD → sab momentum → RSI(14) rakho (simplest, bounded 0-100)
Bollinger, ATR → dono volatility → ATR rakho (stop-loss sizing ke liye useful)
OBV, Volume, VWAP → OBV rakho (cumulative, smart money flow dikhata hai)

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

Result: Decision time avg 8min se → 90 sec ho gayi. Win rate improve hui (52% → 60%) kyunki kam noise → clearer patterns.

KYUN yeh step? EMA(50) trend pakadta hai. RSI KYUN? Overbought/oversold extremes flag karta hai. OBV KYUN? Confirm karta hai ki institutions accumulate kar rahe hain ya distribute. Har ek ek distinct sawaal ka jawaab deta hai.

The Correlation Screening Test

Koi indicator add karne se pehle, uska correlation apne existing set ke saath compute karo.

Formula Correlation Screening Heuristic (rule of thumb, law nahi)

Candidate indicator $I_{new}$ aur existing set ${I_{1}, \dots, I_{n}}$ ke liye:

\overset{ρ}{ˉ}_{new} = \frac{1}{n} i = 1 \sum n ∣ ρ (I_{new}, I_{i}) ∣

Practical guide (yeh thresholds convenience ke liye choose kiye gaye heuristics hain, theory se derive nahi — hamesha backtest karo):

$\overset{ρ}{ˉ}_{new} < 0.5$ → likely new information add karta hai
$0.5 \leq \overset{ρ}{ˉ}_{new} < 0.7$ → marginal value, carefully test karo
$\overset{ρ}{ˉ}_{new} \geq 0.7$ → probably redundant, likely reject karo

KYUN yeh numbers? Yeh roughly shared variance $ρ^{2}$ se map karte hain: $ρ = 0.7$ par, lagbhag aadha variance ( $0.49$ ) shared hai, jise bahut se traders "too much overlap" maante hain. 0.7 mein kuch magic nahi — apne data par validate karo.

Example: Tumhare paas RSI(14) hai. Stochastic add karne ka soch rahe ho.

252 days ke data par $ρ (RSI, Stochastic)$ compute karo → $ρ = 0.87$
Kyunki $0.87 > 0.7$ (shared variance $\approx 0.76$ ) → Stochastic likely reject karo, yeh redundant hai.

Advanced: Indicators ko Portfolio ki Tarah Treat Karo

Indicators ko investments ki tarah treat karo. Return = decision quality improvement. Cost = added complexity/confusion.

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

Heuristic: Ek indicator tumhare net results improve karna chahiye added complexity ki wajah se jo extra decisions tum miss ya delay karoge, uske baad.

Recall 12-saal ke bachche ko explain karo

Socho tum decide karne ki koshish kar rahe ho ki baarish hogi ya nahi. Tumhare paas hai:

Ek thermometer
Ek barometer
Ek hygrometer (humidity)
Ek weather app
Clouds dekh rahe ho
5 padosiyon se pooch rahe ho kya lagte hain
Check kar rahe ho ki tumhara ghutna dard kar raha hai ya nahi (purani injury baarish se pehle takleef deti hai)

Yeh 7 "indicators" hain! Lekin problem yeh hai: tumhare padosi sab usi weather app dekh rahe hain, toh 5 padosiyon se poochna actually app 5 baar check karne jaisa hai. Thermometer aur barometer closely linked hain. Tumhara ghutna aur clouds hi alag clues hain.

Trading mein: Tum jaanna chahte ho ki stock upar jaayega ya nahi. Tum check kar sakte ho ki yeh apne average price se upar hai (trend), tum zyada fast toh move nahi kar raha (momentum), aur bade investors khareed rahe hain ya nahi (volume). Yeh 3 alag clues hain. Lekin agar tum 5 alag "momentum" indicators check karo, tum bas wahi sawaal 5 baar pooch rahe ho. Tum zyada busy feel karte ho, lekin smarter nahi ho—aur ab tum itne busy ho ki baarish (ya stock move) miss kar lete ho kyunki abhi bhi analyze kar rahe ho!

Connections

3.4.1-RSI-Relative-Strength-Index – Core momentum indicator, doosre add karne se pehle ise deeply samjho
3.4.8-Bollinger-Bands – Volatility measure, trend (MA) + volatility (SD) combine karta hai
3.3.2-Moving-Average-Crossovers – Trend system, sirf 2 MAs ke saath complete strategy ho sakti hai
2.5-Trading-Psychology – Analysis paralysis psychological hai, technical nahi
4.2-Position-Sizing – ATR yahan useful hai (stop-loss sizing), entry signals ke liye nahi
5.1-Backtesting-Basics – Test karo ki added indicators actually results improve karte hain ya nahi
3.4.15-Building-a-Trading-System – Agla step: apne 2-3 indicators ko rules mein combine karo

#flashcards/stock-market

Indicator overload kya hai?

Itne zyada technical indicators (typically 4+) use karna ki cognitive load decision quality se tezi se badhti hai, jisse analysis paralysis aur missed opportunities hote hain.

Do correlated indicators ke beech information overlap correctly kaise measure karte hain?

Jointly Gaussian variables ke liye mutual information use karo:

M I = - \frac{1}{2} ln (1 - ρ^{2})

. Yeh logarithmic hai (

ρ \to 1

par infinity par,

ρ \to 0

par 0 par), linear subtraction of

ρ

NAHI.

"Correlation 0.89 = 89% information shared" galat kyun hai?

Kyunki

ρ

linear association measure karta hai, information nahi.

ρ = 0.89

ka matlab hai ~

ρ^{2} = 79%

variance explained, aur information overlap logarithmic hai (

M I \approx 0.39

nats), linearly 89% nahi.

RSI aur Stochastic Oscillator redundant kyun hain?

Dono momentum oscillators hain jo bounded scale par price change ki speed measure karte hain. Correlation typically

ρ > 0.85

hota hai, isliye zyaatar variance (aur information) shared hai.

Optimal indicators ki sankhya kya hai?

Distinct categories (trend, momentum, volume) se 2-3 indicators, jitna ho sake uncorrelated choose kiye gaye.

Redundancy rule of thumb ke liye kaunsa correlation range use hota hai?

∣ ρ ∣ ≳ 0.7

(shared variance

ρ^{2} \approx 0.49

) "too much overlap" ke liye ek common heuristic hai — lekin yeh theoretical law nahi hai; hamesha backtest karo.

Teen distinct indicator categories batao

Trend (e.g., moving averages), Momentum (e.g., RSI), Volume/Volatility (e.g., OBV, ATR)—har ek ek alag market dimension measure karta hai.

Kya

T \propto n^{1.5}

ek proven law hai?

Nahi. Yeh ek illustrative heuristic hai jo linear reading cost (

n

) aur pairwise-comparison cost (

n^{2}

) ke beech interpolate karta hai. Miller's Law working-memory capacity (7±2) ke baare mein hai, time scaling ke baare mein nahi.

T \propto n^{1.5}

ke under, 6× zyada indicators kitne slow hote hain?

6^{1.5} \approx 14.7

×, yani roughly 15× slow — na 6× aur na 30×.

Zyada indicators total wins kyun kam kar sakte hain?

Complexity ki wajah se analysis paralysis trades miss karwa deti hai. Agar added indicators trade frequency 40% kam kar dein lekin win rate sirf 7% improve karein, toh net wins kam ho jaate hain.

Indicator overload ke teen symptoms kya hain?

(1) Analysis paralysis (har setup ke liye >10 min), (2) conflicting signals (>40% time), (3) bahut zyada sources se confirmation ka wait karte hue moves miss karna.