Assignment – Module 4

1. Determine the autocorrelation at lag 3 for the data given below. Check if this correlation

is significant at 95% confidence level.

14,000 17,700 17,500 15,500 20,500 18,100 15,800 14,900 16,300

14,900 17,600 17,000 17,300 18,300 19,100 17,900 19,400 22,900
16,200 14,300

2. For the data given below, estimate the spectral densities for p=1,2 and 3 with notations

followed in the lectures, for a maximum lag of 2

Year 1 2 3 4 5 6 7 8 9 10 11
Peak flow
2160 3210 3070 4000 3830 978 6090 1150 6510 3070 3360
(m3/sec)

3. Statistical properties (in Mm3) of streamflow at a site in the three seasons of a year are

given below

___ ___________________________________
Season I Season II Season III
______________________________________
Mean 35 15 8
Std. Devaition 40 10 6
Lag one
Correlation 0.43 0.67 0.5
______________________________________

The lag one correlation is the correlation of flows with those of the previous season.

Using a Non-stationary, First Order Markov Model, generate streamflow data for 3 years

at the site. State the assumptions you make in using such a model.

4. Given the auto-correlations, r1 = -0.671 and r2 = 0.463, obtain the initial estimates of

parameters of an ARIMA (2,1,0) model using Yule Walker equations.

5. Express the following two models in the form ARIMA (p,d,q):

a) (1-0.2B) (1-B) Xt = ( 1 – 0.5B) et

b) (1-B) Xt = (1 – 0.2B) et

6. Monthly streamflow data for 33 years is analysed. The correlogram (autocorrelation

function), partial autocorrelations and the line spectrum, with usual notations, are shown

in the figure above. (a) Identify (determine) the periodicities in the data, (b) Which model

of the ARMA family is best suited for this data ? Why?

1
Auto-correlation --->

Correlogram
0.5

-0.5
20 0 20 40 60 80 100

Lag --->
Partial Auto-correlation --->

Partial Autocorrelogram
0.5

-0.5
0 10 20 30 40 50 60 70 80 90 100
8
x 10 Lag --->
6
Line Spectrum
4
I --->

0
0 1 2 3 4 5 6 7
w(k) --->
7. Based on the following streamflow statistics (in appropriate units), generate 50 values of

streamflow data using the first order, seasonal, Markov model. Compute and plot the

correlogram and spectral density function for the generated data. ρj is the correlation with

next month, j+1

Month, j xj sj ρj

Oct 5.02 2.31 0.61

Nov 6.50 3.38 0.58
Dec 7.33 3.23 0.50
Jan 6.42 2.95 0.31
Feb 5.35 2.62 0.38
Mar 5.02 1.66 0.37
Apr 6.42 1.80 0.44
May 10.70 2.89 0.34
Jun 12.76 3.32 0.17
Jul 9.05 3.26 0.65
Aug 4.44 1.47 0.93
Sep 3.29 1.22 0.51

8. Monthly streamflows (in Mm3) at a location for the year 2010-11 (upto March 2011) are

given below

Year :2010

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

19 15 39 102 90 29 90 46 30 66 80 89

Year : 2011

Jan Feb Mar

82 17 26

Using a double moving average of order 4, obtain the forecast of streamflow for April

2011.
9. The table below gives a time series composed of 60 values. Plot this series to identify its

trend. Compute the first differences and plot the resulting series. (a)Plot the

autocorrelations upto 15 lags for both the original and the differenced series. Also

determine the first 3 partial auto correlations for the original data. (b) Plot the power

spectrum of the original and the differenced data

Period Observation Period Observation Period Observation

1 9.56 21 60.50 41 85.28
2 12.48 22 63.29 42 84.44
3 13.64 23 66.55 43 86.59
4 18.80 24 68.65 44 88.05
5 25.04 25 72.66 45 90.83
6 30.33 26 71.25 46 93.05
7 34.08 27 65.48 47 94.65
8 40.10 28 62.68 48 96.66
9 42.40 29 56.60 49 96.30
10 41.36 30 49.90 50 96.09
11 39.25 31 49.82 51 99.27
12 38.20 32 51.87 52 104.77
13 41.47 33 57.74 53 105.51
14 46.14 34 58.24 54 105.19
15 52.62 35 58.31 55 109.16
16 59.01 36 59.91 56 110.78
17 60.20 37 62.61 57 115.77
18 58.53 38 69.07 58 122.75
19 56.98 39 77.36 59 126.85
20 57.82 40 80.39 60 132.57
10. The inflow data (in million cubic feet) to a reservoir is given below

Year Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May
1940-
41 29,243 87,671 77,982 14,465 12,527 23,769 5,976 590 2,391 1,245 1,339 3,039

41-42 29,019 76,623 56,931 18,032 13,877 2,633 3,616 580 731 258 792 1,644

42-43 14,916 97,800 56,286 24,710 13,262 3,714 3,957 2,253 897 588 785 6,690

43-44 17,800 107,200 24,321 29,113 38,651 12,887 3,717 1,781 1,264 1,148 610 1,851

44-45 4,665 78,799 34,444 9,453 12,560 8,985 4,078 1,816 959 562 836 950
1945-
46 5,860 71,985 31,280 17,474 12,322 4,454 2,153 1,446 1,162 925 1,006 996

46-47 20,720 77,108 113,153 27,862 14,683 16,006 6,914 1,259 1,068 766 1,321 1,112

47-48 1,745 55,929 58,237 39,271 21,552 2,236 3,063 1,548 952 275 1,759 3,725

48-49 16,421 55,903 96,601 23,747 15,347 7,138 4,026 1,622 841 727 731 1,870

49-50 13,942 44,977 51,101 24,715 15,400 4,822 2,727 1,831 1,346 482 78 789
1950-
51 7,855 91,462 44,170 49,455 14,721 6,409 2,176 1,608 920 432 1,157 2,087

51-52 15,052 55,179 44,183 16,420 20,326 5,104 2,321 722 952 609 623 1,050

52-53 9,430 36,096 55,821 10,194 24,359 2,914 4,109 1,701 865 720 696 260

53-54 9,658 104,243 100,664 15,400 35,493 3,509 1,446 677 198 168 603 4,314

54-55 20,101 80,172 74,683 22,160 22,692 2,133 3,549 1,718 610 248 879 8,858
1955-
56 23,448 26,032 25,025 27,574 29,877 8,225 4,114 2,030 1,163 584 848 2,748

56-57 28,459 87,026 57,194 14,943 27,650 20,217 2,463 341 529 535 469 5,178

57-58 16,196 83,511 41,085 9,148 11,072 14,093 3,159 1,469 1,059 634 1,493 5,670

58-59 14,394 131,338 51,913 31,080 18,479 7,267 3,441 1,334 896 304 688 1,552

59-60 26,616 196,988 54,466 52,557 13,280 6,893 3,481 1,337 582 818 1,988 3,325
Develop candidate ARIMA models and select the best model for the data series based on

following methods.

– Maximum likelihood (ML)

– Minimum mean square error (MMSE)

Validate the models using following tests

– Significance of residual mean

– Significance of periodicities

– Cumulative periodogram test or Bartlett’s test

– White noise test

– Whittle’s test

– Portmanteau test