Cheat Sheet Time Series Analysis

Created by Daniel Zaldaña @ZaldanaDaniel

Time Series Analysis

Cheat Sheet

1. Basic Concepts 3. Autocorrelation

Time Series Components Correlation Measures

• Trend (Tt ): Long-term direction ACF:

•
Pn
Seasonal (St ): Regular patterns
t=k+1 (Xt − X̄)(Xt−k − X̄)
• Cyclical (Ct ): Irregular fluctuations ρk = Pn 2
t=1 (Xt − X̄)
• Random (Rt ): Unexplained variation

PACF:

ϕkk = Corr(Xt , Xt−k |Xt−1 , ..., Xt−k+1 )

Decomposition Models

Additive: Yt = Tt + St + Ct + Rt

Multiplicative: Yt = Tt × St × Ct × Rt 4. Time Series Models

Model Specifications
AR(p):
p
2. Statistical Measures Xt = c +
X
ϕi Xt−i + εt
i=1

Core Statistics MA(q):

Mean: q
n
1X X
X̄ = Xt Xt = µ + εt + θi εt−i
n t=1 i=1

Moving Average: ARMA(p,q):

k−1 p q
1X X X
M A(k) = Xt−i Xt = c + ϕi Xt−i + θi εt−i + εt
k i=0
i=1 i=1

Variance: ARIMA(p,d,q):
n
1 X p q
σ2 = (Xt − X̄)2 X X
n−1 (1 − B)d (Xt − µ) = ϕi Xt−i + θi εt−i + εt
t=1
i=1 i=1
Cheat Sheet Time Series Analysis

5. Forecasting Metrics 8. Time Series Patterns

Error Measures Pattern Analysis

MAE: Exponential Growth: Yt = Y0 (1 + r)t
n K
1X Logistic Growth: Yt =
|Yt − Yˆt |
K−Y0 −rt
M AE = 1+( Y0 )e
n t=1 Holt-Winters Multiplicative:
Yt
RMSE: Level: Lt = α + (1 − α)(Lt−1 + Tt−1 )
St−s
v
u n
u1 X Trend: Tt = β(Lt − Lt−1 ) + (1 − β)Tt−1
RM SE = t (Yt − Yˆt )2 Yt
n t=1 Seasonal: St = γ + (1 − γ)St−s
Lt

MAPE:
n
100% X Yt − Yˆt 9. Spectral Analysis
M AP E = | |
n t=1 Yt
Frequency Domain
1
Pn −iωt 2
Periodogram: I(ω) = 2πn | t=1
P∞Xt e |
Spectral Density: f (ω) = 2π k=−∞ γk e−ikω
1

6. Stationarity Tests Coherence: Cxy (ω) =

|fxy (ω)|2
fxx (ω)fyy (ω)

Stationarity Conditions
Weak Stationarity: 10. Model Diagnostics
• E[Xt ] = µ (constant mean)
• V ar(Xt ) = σ 2 (constant variance) Diagnostic Tests
• Cov(Xt , Xt+k ) = γk (covariance depends only Pn
(e −et−1 )2
Durbin-Watson: DW = Pnt
t=2
2
on k) t=1 et
Breusch-Godfrey: LM = (n − p)R2
Augmented Forecast Error Decomposition:
Pp Dickey-Fuller: ∆yt = α + βt +
γyt−1 + i=1 δi ∆yt−i + εt
1X
PT
St2 Pt M SE = (Yt − Yˆt )2
KPSS Test: ηc = Tt=1 where n
2 σ̂ 2
PpSt = i i=1 ei ¯
Unit Root Condition: |1 − i=1 ϕi z | ̸= 0 for = (Ȳ − Ŷ )2 + (sY − rsŶ )2 + 2(1 − r)sY sŶ
|z| ≤ 1

11. Seasonal Analysis

7. Advanced Statistical Measures
Seasonal Metrics
Additional Metrics Seasonal Index:
Cross-Correlation Function (CCF): ρxy (k) = Average for season i
SIi = × 100
Pn
t=k+1 (Xt −X̄)(Yt−k −Ȳ )
√Pn 2
√Pn 2
Overall average
t=1 (Xt −X̄) t=1 (Yt −Ȳ )
Ljung-Box Q-statistic: Q(m) = n(n +
Pm ρ̂2k
2) k=1 n−k SARIMA Model:
Information Criteria: AIC = −2 ln(L) + 2k
BIC = −2 ln(L) + k ln(n) ϕp (B)ΦP (B s )(1−B)d (1−B s )D yt = θq (B)ΘQ (B s )εt
Cheat Sheet Time Series Analysis

1. Basic Statistics
Statistical Components

1 import numpy as np
2 import pandas as pd
3
4 def c a l c u l a t e _ b a s i c _ s t a t s ( data ) :
5 """ Calculate basic statistics """
6 return {
7 ’ mean ’: np . mean ( data ) ,
8 ’ variance ’: np . var ( data , ddof =1) ,
9 ’ std ’: np . std ( data , ddof =1)
10 }
11
12 def moving_average ( data , window ) :
13 """ Calculate moving average """
14 return pd . Series ( data ) . rolling (
15 window = window
16 ) . mean () . values

2. Correlation Analysis
ACF and PACF

1 from statsmodels . tsa . stattools import (
2 acf , pacf
3 )
4
5 def c o r r e l a t i o n _ a n a l y s i s ( data , lags ) :
6 """ Calculate ACF and PACF """
7 acf_vals = acf (
8 data ,
9 nlags = lags ,
10 fft = True
11 )
12 pacf_vals = pacf (
13 data ,
14 nlags = lags ,
15 method = ’ yw ’
16 )
17 return {
18 ’ acf ’: acf_vals ,
19 ’ pacf ’: pacf_vals
20 }
Cheat Sheet Time Series Analysis

3. Time Series Models

Model Implementations

1 from statsmodels . tsa . arima . model import (
2 ARIMA
3 )
4
5 def fit_arima_models ( data ) :
6 """ Fit ARIMA and variants """
7 models = {}
8

9 # AR (1) model
10 models [ ’ ar ’] = ARIMA (
11 data ,
12 order =(1 , 0 , 0)
13 ) . fit ()
14

15 # MA (1) model
16 models [ ’ ma ’] = ARIMA (
17 data ,
18 order =(0 , 0 , 1)
19 ) . fit ()
20

21 # ARMA (1 ,1)
22 models [ ’ arma ’] = ARIMA (
23 data ,
24 order =(1 , 0 , 1)
25 ) . fit ()
26

27 # ARIMA (1 ,1 ,1)
28 models [ ’ arima ’] = ARIMA (
29 data ,
30 order =(1 , 1 , 1)
31 ) . fit ()
32

33 return models
Cheat Sheet Time Series Analysis

4. Forecasting Metrics
Error Measures

1 def c alcula te_met rics ( actual , pred ) :
2 """ Calculate forecast metrics """
3

4 # Mean Absolute Error

5 mae = np . mean ( np . abs (
6 actual - pred
7 ))
8

9 # Root Mean Square Error

10 rmse = np . sqrt ( np . mean (
11 ( actual - pred ) ** 2
12 ))
13
14 # Mean Absolute Percentage Error
15 mape = np . mean ( np . abs (
16 ( actual - pred ) / actual
17 ) ) * 100
18
19 return {
20 ’ mae ’: mae ,
21 ’ rmse ’: rmse ,
22 ’ mape ’: mape
23 }
Cheat Sheet Time Series Analysis

5. Stationarity Tests
Stationarity Analysis

1 from statsmodels . tsa . stattools import (
2 adfuller , kpss
3 )
4
5 def ch ec k_ st ati on ar it y ( data ) :
6 """ Perform stationarity tests """
7
8 # ADF Test
9 adf_result = adfuller ( data )
10
11 # KPSS Test
12 kpss_result = kpss ( data )
13
14 # Rolling statistics
15 roll_mean = pd . Series ( data
16 ) . rolling ( window =12) . mean ()
17 roll_std = pd . Series ( data
18 ) . rolling ( window =12) . std ()
19
20 return {
21 ’ adf_stat ’: adf_result [0] ,
22 ’ adf_pval ’: adf_result [1] ,
23 ’ kpss_stat ’: kpss_result [0] ,
24 ’ kpss_pval ’: kpss_result [1] ,
25 ’ roll_mean ’: roll_mean ,
26 ’ roll_std ’: roll_std
27 }
Cheat Sheet Time Series Analysis

6. Advanced Measures
Advanced Statistics

1 from statsmodels . stats . diagnostic import (
2 acorr_ljungbox
3 )
4
5 def advanced_stats ( data ) :
6 """ Calculate advanced metrics """
7
8 # Ljung - Box Test
9 lb_test = acorr_ljungbox (
10 data ,
11 lags =10
12 )
13
14 # Information Criteria
15 model = ARIMA (
16 data ,
17 order =(1 ,0 ,0)
18 ) . fit ()
19
20 return {
21 ’ lb_stat ’: lb_test . lb_stat ,
22 ’ lb_pval ’: lb_test . lb_pvalue ,
23 ’ aic ’: model . aic ,
24 ’ bic ’: model . bic
25 }
Cheat Sheet Time Series Analysis

7. Seasonal Components
Seasonal Analysis

1 from statsmodels . tsa . seasonal import (
2 se as on al _d ec om po se
3 )
4
5 def a na l y z e_ s e as o n al i t y ( data , period ) :
6 """ Analyze seasonal patterns """
7
8 # Decomposition
9 decomp = s ea so na l_ de co mp os e (
10 data ,
11 period = period ,
12 model = ’ multiplicative ’
13 )
14

15 # Seasonal Indices
16 indices = pd . Series (
17 decomp . seasonal
18 ) . unique ()
19
20 return {
21 ’ trend ’: decomp . trend ,
22 ’ seasonal ’: decomp . seasonal ,
23 ’ residual ’: decomp . resid ,
24 ’ indices ’: indices
25 }
Cheat Sheet Time Series Analysis

8. Growth Models
Growth Patterns

1 from scipy . optimize import curve_fit
2

3 def f it_gro wth_mo dels ( data , time ) :

4 """ Fit growth models """
5
6 def exp_growth (t , y0 , r ) :
7 return y0 * (1 + r ) ** t
8

9 def logistic (t , K , y0 , r ) :
10 return K / (1 + (
11 ( K - y0 ) / y0
12 ) * np . exp ( - r * t ) )
13
14 # Fit models
15 exp_params = curve_fit (
16 exp_growth ,
17 time ,
18 data
19 ) [0]
20

21 log_params = curve_fit (
22 logistic ,
23 time ,
24 data
25 ) [0]
26

27 return {
28 ’ exp ’: exp_params ,
29 ’ logistic ’: log_params
30 }
Cheat Sheet Time Series Analysis

9. Spectral Analysis
Frequency Domain

1 from scipy import signal
2

3 def s pectra l_anal ysis ( data , fs =1.0) :

4 """ Perform spectral analysis """
5
6 # Periodogram
7 freqs , psd = signal . periodogram (
8 data ,
9 fs = fs
10 )
11
12 # Spectral Density
13 f_welch , psd_welch = signal . welch (
14 data ,
15 fs = fs
16 )
17
18 return {
19 ’ freq ’: freqs ,
20 ’ psd ’: psd ,
21 ’ welch_freq ’: f_welch ,
22 ’ welch_psd ’: psd_welch
23 }

10. Model Diagnostics

Diagnostic Tests

1 from statsmodels . stats . diagnostic import (
2 durbin_watson ,
3 acorr_breusch_godfrey
4 )
5
6 def m odel_d iagnos tics ( model , resid ) :
7 """ Perform model diagnostics """
8

9 # Durbin - Watson
10 dw = durbin_watson ( resid )
11
12 # Breusch - Godfrey
13 bg = a c or r _ b r e u s c h _ g o d f r e y (
14 model ,
15 nlags =5
16 )
17
18 return {
19 ’ dw_stat ’: dw ,
20 ’ bg_stat ’: bg [0] ,
21 ’ bg_pval ’: bg [1]
22 }