Descriptive Statistics
N Minimum Maximum Mean Std. Deviation
SP500 1833 2.7300 4674.7727 376.970841 808.2854263
Dividend 1833 .0000 68.7100 7.469520 13.6383099
Earnings 1833 .0000 197.9100 17.554663 34.5599588
Consumer Price Index 1833 6.2800 306.1300 65.328385 80.1370175
Long Interest Rate 1833 .6200 15.3200 4.487376 2.2976223
Real Dividend 1833 .0000 71.2200 20.212477 13.5273637
Real Earnings 1833 .0000 217.2600 41.105701 37.6597317
Real Price 1833 82.8700 5140.5900 743.788865 915.5151930
Valid N (listwise) 1833
The descriptive tables show that high standard deviations (e.g., SP500, Real Price, Earnings)
indicate significant market fluctuations. Increasing values over time suggest the need for trend
analysis (e.g., time series decomposition, regression). Strong inflation impact (CPI, Real Price,
Real Earnings) implies that a real vs. nominal analysis could be useful.
Correlation analysis (Pearson correlation)
This is used to measure the strength and direction of the relationship between S & P 500 index
and technology-driven- stock prices or sector indices. Using SPSS, correlation analysis was done
to show the relationship and strength of the two variables, below is the output.
Correlations
SP500 PE10 Real Price Earnings
Pearson Correlation 1 .572** .976** .948**
SP500 Sig. (2-tailed) .000 .000 .000
N 1833 1833 1833 1833
Pearson Correlation .572** 1 .673** .532**
PE10 Sig. (2-tailed) .000 .000 .000
N 1833 1833 1833 1833
Pearson Correlation .976** .673** 1 .932**
Real Price Sig. (2-tailed) .000 .000 .000
N 1833 1833 1833 1833
Pearson Correlation .948** .532** .932** 1
Earnings Sig. (2-tailed) .000 .000 .000
N 1833 1833 1833 1833
**. Correlation is significant at the 0.01 level (2-tailed).
To check on the other variables, intercorrelation was done for all the variables: real price,
earnings, real Dividend, real earnings, long interest rate, consumer price index, and divide. The
output below shows how all the variables interrelate and the direction and strength of their
relationship.
Model Description
Model Type
Consumer Price IndexModel_1 ARIMA (0,0,0)
Model ID
Earnings Model_2 ARIMA (0,0,0)
ARIMA (0,0,0) is equivalent to a simple mean model, meaning the best predictor of future
values is just the average of past values. No autoregressive (AR) terms, differencing (I), or
moving average (MA) terms were used. ARIMA (0,0,0) on CPI (Model_1). This suggests that CPI
is not strongly time-dependent and does not follow a clear trend or seasonality. A more
appropriate model might include an AR or MA component if the data shows autocorrelation.
ARIMA (0,0,0) on Earnings (Model_2). This indicates that past earnings do not strongly predict
future earnings. If earnings have a trend, a differencing term (I=1) may be necessary.
Model Fit
Fit Mean SE Minimu Maximu Percentile
Statistic m m
5 10 25 50 75 90 95
Stationary
.839 .085 .780 .899 .780 .780 .780 .839 .899 .899 .899
R-squared
R-squared.839 .085 .780 .899 .780 .780 .780 .839 .899 .899 .899
RMSE 24.301 18.864 10.962 37.640 10.962 10.962 10.962 24.301 37.640 37.640 37.640
170.81 121.94 121.94 121.94 170.81
MAPE 69.114 121.940 219.682 219.682219.682219.682
1 0 0 0 1
923.86 711.10 1426.68 421.03 421.03 421.03 923.86 1426.68 1426.68 1426.68
MaxAPE 421.034
2 6 9 4 4 4 2 9 9 9
MAE 16.998 17.291 4.772 29.225 4.772 4.772 4.772 16.998 29.225 29.225 29.225
174.16 162.93 162.93 162.93 174.16
MaxAE 15.873 162.937 185.385 185.385185.385185.385
1 7 7 7 1
Normalize
6.031 1.745 4.797 7.264 4.797 4.797 4.797 6.031 7.264 7.264 7.264
d BIC
From the above table for testing the goodness of fit: The R-squared (0.839) is strong, but the
MAPE (170.811%) is very high, meaning the model is not accurate for forecasting. A large Max
APE (923.862%) suggests that the model struggles with some extreme values.
Model Statistics
Model Number ofModel FitLjung-Box Q(18) Number of
Predictors Statistics Outliers
Stationary R-Statistics DF Sig.
squared
Consumer Price Index-
1 .780 29618.144 18 .000 0
Model_1
Earnings-Model_2 1 .899 4458.219 18 .000 0
Checking the correlation between technology stock returns and S&P 500 returns over the past
decade. This is done in two time periods. Then we compare these correlations to see if they
have decreased. Statistically, this is done using Fisher’s Z-test for correlation comparison. To
obtain this we first compute monthly returns for S&P 500 and Technology Stocks as follows
Monthly return = (current month value -previous month value)/ (previous month value). Using
statistical software, we create these two variables; sp500-return and technology stock returns.
Since we are working on decades, we split the data to get the past decade. Checking the
correlation between the two, we have the table below
Correlations in the first decade of 2002-2012
SP500RETURN TechStock_Ret
urn
Pearson Correlation 1 .995**
SP500RETURN Sig. (2-tailed) .000
N 130 130
Pearson Correlation .995** 1
TechStock_Return Sig. (2-tailed) .000
N 130 130
**. Correlation is significant at the 0.01 level (2-tailed).
Apply Fisher's Z-transformation formula
Z= ½ln((1+r/1-r))
substituting r in the formula we have,
Z =1/2ln (1+0.95)/ (1-0.95))
This gives,
Z=2.9945
The Fisher's Z-transformed value for the correlation r=0.995r = 0.995r=0.995 is approximately
2.9945
Correlations second decade, 2013-2023
TechStock_Ret SP500Returns
urn
Pearson Correlation 1 .994**
TechStock_Return Sig. (2-tailed) .000
N 127 127
Pearson Correlation .994** 1
SP500Returns Sig. (2-tailed) .000
N 127 127
**. Correlation is significant at the 0.01 level (2-tailed).
To obtain the comparison, we calculate using Z Fisher’s transformation. Applying the formula we
have;
Apply Fisher's Z-transformation formula
Z= ½ln((1+r/1-r))
substituting r in the formula we have,
Z =1/2ln (1+0.94)/ (1-0.94))
Z= 2.9031
Since the p-value is less than 0.05 for both decades, this shows that the correlation has not
changed significantly over the decade.
Seasonal decomposition
Seasonal decomposition, year-wise and quarterly is useful to understand the stock market.
Model Description
Model Name MOD_2
Model Type Additive
Series Name 1 SP500
Length of Seasonal Period 4
Span equal to the
Computing Method of Movingperiodicity and all
Averages points weighted
equally
Applying the model specifications from MOD_2
From the above table, it is evident that. The trend component shows that the long-term
movement of the S&P 500, filters out short-term fluctuations. For a graphical representation, if
the trend is upward, then the index has been increasing. The seasonal component represents
the repeating short-term fluctuations of each 4-period cycle. For a strong seasonality it it
implies S&P 500 performance fluctuates predictably every quarter.
Seasonal Factors
Series Name: SP500
Period Seasonal
Factor
1 -.0526896
2 -1.2470502
3 -.2029824
4 1.5027222
The above figure is the representation of 4 quarters.
The first period of -0.0527 shows a slightly below-the-average level, -1.24 shows a strong
seasonal decline, -0.2029 shows a moderate seasonal decline and finally, 1.5027 shows a strong
seasonal increase. The S&P 500 tends to rise sharply in Q4, likely due to year-end rallies,
earnings seasons, or market optimism.
Removing this seasonality to obtain a true trend, we perform an ANOVA table.
ANOVA
QUARTER, period 4
Sum ofdf Mean Square F Sig.
Squares
Between Groups 2.249 458 .005 .003 1.000
Within Groups 2290.000 1374 1.667
Total 2292.249 1832
F-value (0.003). This is extremely low, meaning the difference between groups is insignificant.
p-value (Sig. = 1.000). A p-value of 1.000 means no statistically significant difference between
groups. In other words, quarterly variations do not significantly impact the dependent variable.
Between-Groups vs. Within-Groups Variability. The between-groups mean square (0.005) is
much smaller than the within-groups mean square (1.667). This implies that quarterly periods
(Period = 4) do not significantly impact the dependent variable. If you expected seasonality, the
ARIMA model might need a different seasonal structure.
Non-periodic nature: The label "YEAR, not periodic" suggests that the data lacks a repeating
pattern. If you're testing for seasonality, the trend should exhibit cyclical behavior rather than a
constant line. data compression issue: The black horizontal band suggests that many data points
are overlapping at the same value (around 2.5). This could indicate a problem with how the data
is structured or aggregated. Possible issue with period selection: If seasonality exists, it may not
be best captured using a period of 4 (quarterly). You might need to test monthly (12), weekly
(52), or another relevant cycle.
Consumer Price Index (CPI) Trend. The CPI appears to be following a steady upward trend with
an exponential increase in later years. This suggests inflationary effects over time.
Earnings Trend. Earnings remained relatively low before experiencing a sharp increase followed
by high volatility towards the end. The extreme spikes and drops at the end could indicate data
issues (e.g., missing values or structural breaks).
Coefficients
Model Unstandardized CoefficientsStandardized t Sig. Collinearity Statistics
Coefficients
B Std. Error Beta Tolerance VIF
(Constant) 326.918 5.106 64.028 .000
1
SP500 1.106 .006 .976 193.120 .000 1.000 1.000
a. Dependent Variable: Real Price
Regression Equation:
Real Price=326.918+1.106×SP500\text {Real Price} = 326.918 + 1.106 {SP500}
Real Price=326.918+1.106×SP500. The constant (intercept) is 326.918, meaning when SP500 =
0, the predicted Real Price is 326.918. The coefficient of SP500 is 1.106, meaning that for every
one-unit increase in SP500, the Real Price increases by 1.106.
Significance Testing: t-value for SP500 = 193.120, with a p-value of 0.000. This suggests a highly
significant relationship between SP500 and Real Price.
Collinearity Statistics: Tolerance = 1.000, VIF = 1.000 → No multicollinearity issues.
Standardized Beta Coefficient: Beta = 0.976, meaning SP500 explains 97.6% of the variation in
Real Price. This indicates a strong positive correlation between SP500 and Real Price.
From the above prediction results, it is evident that;
The S&P 500 is a strong predictor of Real Price. The model is statistically significant, and SP500
has a strong positive effect on Real Price. No multicollinearity concerns (since VIF < 10 and
Tolerance > 0.1).