# Before determining the impact of macroeconomic factors on the current account deficit

The current account balance is an important macroeconomic factor. Though many countries similar to India are spending more abroad resulting in a current account deficit, due to which many trade unions, businesses, and parliamentarians are increasingly concerned with foreign expenditures. As the current account deficit represents a country’s net borrowings, there is an increased preference among investors towards countries that are making policies to reduce the deficit.

In the previous article, the analysis identified that over the years, the value of the current account deficit witnessed many fluctuations. But, in recent times the existence of a trade imbalance resulted in a rising deficit. However, as the trend examination fails to provide information about the factors constituting towards current account deficit, thus, there is a need for statistical examinations.

Statistical analysis is the method that enables linkages between variables by identifying the macroeconomic factors constituting towards current account deficit. However, as the relationship could only be drawn once the efficiency of the linkage is verified, this article aims to assess the model to verify the impact of macroeconomic factors on the current account deficit of India.

## Assumption testing to establish consistency of macroeconomic data

In statistical analysis, all parametric tests assume certain characteristics about the data, known as assumptions. Testing these assumptions ensures that the results that are obtained from the analysis of the macroeconomic dataset are viable and consistent. Any change in the assumptions can alter the results of the analysis. Generally, an OLS regression assumes that the data needs to be stationary, homoscedastic, non-collinear, and normal, thus, herein all the assumptions are tested.

### Stationarity check using Dickey-Fuller test on macroeconomic data to establish consistency of data over time

Stationarity can be mathematically defined as a measure of some stochastic process where the cumulative distribution function describing this process doesn’t depend on a time parameter (Hopkins, 2019). In simpler words, it verifies if the statistical properties of a time series, such as the mean and variance, are constant over time.

Generally, stationarity is determined by running the Dickey-Fuller test. This test is run on the dataset where the correlation in the error term is checked for correlation. The results of the analysis for the current account deficit model are shown below:

Variable | Level | First Difference | Second Difference |
---|---|---|---|

Current Account Balance | 0.43 | 0.00 | – |

GDP | 1.00 | 0.00 | – |

Private Investment (%GDP) | 0.26 | 0.00 | – |

Government Expenditure(%GDP) | 0.66 | 0.00 | – |

Trade Openness | 0.57 | 0.00 | – |

Trade Balance | 0.63 | 0.00 | – |

External Debt | 0.33 | 0.00 | – |

Fiscal Deficit | 0.99 | 0.00 | – |

FDI Net Inflows | 0.74 | 0.00 | – |

Foreign Exchange Reserve | 0.55 | 0.00 | – |

Broad Money Supply | 1.00 | 0.99 | 0.00 |

Rate of inflation | 0.10 | 0.00 | – |

Rate of Interest | 0.09 | 0.00 | – |

Exchange Rate | 0.97 | 0.00 | – |

To interpret the results of the Dickey-Fuller test, the Z(t) value and Mackinnon approximate p-value for Z(t) are to be considered. If the Z(t) is not a large negative number and if the p-value is not at least on a 5% significance level, then the data is non-stationary (Xiao, 2001).

Due to the dataset being non-stationary, further analysis can’t be performed on it. Herein, the results depict that for all the variables the p-value at the level test is more than 0.05. Thus, the variables are not stationary. For this, the 1^{st} difference for each of the variables is computed. As herein, for all variables except broad money supply, the p-value is less than 0.05, thus, the null hypothesis of having a unit root in the dataset is rejected representing the presence of stationarity. Lastly, for broad money supply at the 2^{nd} difference, the p-value is 0.00, 0.05, thus the variable is stationary at the 2^{nd} difference. Hence, all the variable’s stationary form is derived and they could be used for further analysis.

### Normality

The normality test helps in assessing the probability that a random variable underlying the data set is distributed across the sample mean. As most of the parametric datasets need consistent data for linkage development, there is a requirement to assess the nature of the distribution (Singh & Masuku, 2021). Out of the different tests, the skewness and kurtosis test is also the most commonly used method for checking normality. The results of the test are given below:

Variable | Obs | Pr(Skewness) | Pr(Kurtosis) | Adj chi2(2) | Prob>chi2 |
---|---|---|---|---|---|

resid | 25 | 0.10 | 0.50 | 3.50 | 0.17 |

It is seen that for 25 observations, the probability of skewness (Pr(Skewness)) is 0.10 > 0.05. This implies that the skewness is not asymptotically normally distributed. Similarly, the probability of Kurtosis (Pr(Kurtosis)) is 0.50 > 0.05. This means that the Kurtosis is also not asymptotically normally distributed. The Prob>chi2 value is 0.17, which shows the significance of the dataset at a 5% level. As a consequence, the null hypothesis of having a normal distribution can’t be rejected. Thus, the residuals show a normal distribution.

### Heteroskedasticity

Heteroscedasticity occurs when the standard deviations of a predicted variable, monitored over different values of an independent variable or as related to prior periods, are non-constant (Hayes, 2022). The test contributes to determining when the built model is unable to explain the variation in the pattern of the dependent variable due to residuals. As the existence of this variation would result in the existence of an unstable and inefficient regression model, there is a need for constant error variance across the residuals. Herein, for detecting heteroscedasticity in the current account deficit model, the Breusch-Pagan test is conducted. The test for tests are shown below:

Chi-square | Prob |
---|---|

5.38 | 0.02 |

The test finds the probability value of the chi-square statistic to be less than 0.05. Thus, the null hypothesis of Homoscedasticity can be rejected and this implies the presence of heteroscedasticity in the residuals. To resolve the issue, a robust command is used in the regression. The robust standard errors are obtained by adding the ‘vce(robust)’ command after running a regression in STATA. Thus, with the inclusion of robust standard errors, the issue of heteroscedasticity is resolved.

### Multicollinearity

In multicollinearity, the regression model consists of several correlated independent variables. As the inclusion of correlated variables reduces the efficiency of the model and states that non-required variables are included in the model, there is a need to assess multicollinearity (Shrestha, 2020). The multicollinearity for the current account balance model is assessed using VIF, the results of which are shown below:

Variable | VIF |
---|---|

DED | 129.25 |

DFE | 46.25 |

DTB | 26.91 |

DIE | 10.74 |

DGE | 5.68 |

DDBM | 5.47 |

DTO | 4.78 |

DGDP | 4.44 |

DPI | 3.26 |

DFD | 2.71 |

DI | 2.55 |

DIn | 2.50 |

DFDI | 2.37 |

It is maintained as a rule of thumb that VIF values less than 10 imply no multicollinearity (Kim, 2019). Herein, the above table signifies that for initial variables i.e. DED, DFE, DTB, and DIE the value is more than 10. Thus, there is the existence of multicollinearity in the model. As the value of DED is very high, it is eliminated from the model and then the regression model is re-built to assess multicollinearity. The results obtained are:

Variable | VIF |
---|---|

DTB | 8.73 |

DFE | 7.47 |

DIE | 5.51 |

DGE | 4.35 |

DTO | 4.31 |

DGDP | 3.82 |

DI | 2.54 |

DIn | 2.46 |

DPI | 2.28 |

DFD | 2.24 |

DFDI | 2.20 |

DDBM | 2.06 |

Here, it is found that the variables no longer have multicollinearity as the value of VIF is less than 10. Thus, all the mentioned variables could be used for further testing.

### Autocorrelation

Autocorrelation is a mathematical representation of the degree of similarity between a given time series and a lagged version of itself over successive time intervals (Smith, 2023). As the existence of correlation reduces the efficiency of the model, there is a need to have no autocorrelation in the model. The Durbin-Watson test is done to assess the autocorrelation in the model.

Durbin-Watson d-statistic | dL | 4-dU | dU | 4-dL |
---|---|---|---|---|

2.57 | 0.34 | 1.02 | 2.98 | 3.67 |

In the above table, as the Durbin-Watson statistic value lies between 4-dU and dU, the model has no serial correlation among the variables. Hence, all the mentioned variables could be used for impact examination.

## Macroeconomic factors are relevant for assessing the current account balance

The macroeconomic data though is a means of understanding the economic condition but as the data is derived from an open source, there is the existence of non-uniformity in the data. This results in hampering the relationship building between variables and reducing the effectiveness of the model in determining factors’ impact. Herein, as the model needs to be developed for identifying factors impacting the current account balance, thus, initial assumption testing was required. The examination of Autocorrelation, Heteroscedasticity, Stationarity, Normality, and Multicollinearity test determined that external debt is not a relevant variable for impact examination.

#### References

- Hayes, A. (2022).
*Heteroscedasticity Definition: Simple Meaning and Types Explained*. Investopedia. https://www.investopedia.com/terms/h/heteroskedasticity.asp - Hopkins, I. (2019).
*What Is Stationarity?*Medium. https://medium.com/analytics-vidhya/what-is-stationarity-8f16cdfa7e5f - Kim, J. H. (2019). Multicollinearity and misleading statistical results.
*Korean Journal of Anesthesiology*,*72*(6), 558–569. - Shrestha, N. (2020). Detecting Multicollinearity in Regression Analysis.
*American Journal of Applied Mathematics and Statistics*,*June*, 1–5. https://doi.org/10.12691/ajams-8-2-1 - Singh, A. S., & Masuku, M. (2021). Assumption and testing of normality for statistical analysis.
*American Journal of Mathematics and Mathematical Sciences*,*3*(1), 169–175. - Smith, T. (2023).
*Autocorrelation: What It Is, How It Works, Tests*. Investopedia. https://www.investopedia.com/terms/a/autocorrelation.asp - Xiao, Z. (2001). Testing the null hypothesis of stationarity against an autoregressive unit root alternative.
*Journal of Time Series Analysis*,*22*(1).

I am a Senior Analyst at Project Guru, a research and analytics firm based in Gurugram since 2012. I hold a master’s degree in economics from Amity University (2019). Over 4 years, I have worked on worked on various research projects using a range of research tools like SPSS, STATA, VOSViewer, Python, EVIEWS, and NVIVO. My core strength lies in data analysis related to Economics, Accounting, and Financial Management fields.

## Discuss