I have sample dataset, i want to predict following result for 2 periods. But prediction function gives me same results.
This is my dataset (data['t1']);
0 83.846
1 73.350
2 66.499
3 63.576
4 66.545
5 57.264
6 63.009
7 59.608
8 62.775
9 58.451
10 80.893
11 58.734
12 77.830
13 73.374
14 61.650
15 52.548
16 31.683
17 57.599
18 70.814
19 65.354
20 60.033
21 50.162
22 60.764
23 53.799
24 67.266
25 65.520
26 71.248
27 60.457
28 52.424
29 55.622
30 78.149
31 72.111
Code ;
from statsmodels.tsa.arima_model import ARIMA
import pmdarima as pm
model = pm.auto_arima(data['t1'], start_p=1, start_q=1,
test='adf', # use adftest to find optimal 'd'
max_p=5, max_q=5, # maximum p and q
m=1, # frequency of series
d=None, # let model determine 'd'
seasonal=True,
start_P=0,
D=0,
trace=True,
error_action='ignore',
suppress_warnings=True,
stepwise=True)
print(model.summary())
Prediction;
predict, conf_int = model.predict(2,return_conf_int=True,alpha=0.05)
predict
Result ;
array([71.88338364, 71.88338364])
How can i solve this problem? Does something wrong on my auto_arima model?
fit_summary;
Best model: ARIMA(0,1,1)(0,0,0)[0]
Total fit time: 0.579 seconds
SARIMAX Results
==============================================================================
Dep. Variable: y No. Observations: 34
Model: SARIMAX(0, 1, 1) Log Likelihood -126.062
Date: Mon, 15 Nov 2021 AIC 256.124
Time: 16:25:30 BIC 259.117
Sample: 0 HQIC 257.131
- 34
Covariance Type: opg
==============================================================================
coef std err z P>|z| [0.025 0.975]
------------------------------------------------------------------------------
ma.L1 -0.5351 0.156 -3.438 0.001 -0.840 -0.230
sigma2 120.5502 31.181 3.866 0.000 59.436 181.664
===================================================================================
Ljung-Box (L1) (Q): 0.62 Jarque-Bera (JB): 0.01
Prob(Q): 0.43 Prob(JB): 1.00
Heteroskedasticity (H): 1.11 Skew: -0.02
Prob(H) (two-sided): 0.87 Kurtosis: 2.94
===================================================================================
Your ARIMA model only uses the last component, so it is an MA model. Such an MA model can only predict q steps into the future, so in your case only one step. If you want to predict more than one step, you either need to increase q or switch to an AR model.