MAKE MOST OF THE KNOWLEDGE NETWORK, JOIN ACADEMIC RESEARCH FOUNDATION

This study evaluates the performance of machine learning-based IV methods-specifically Double Machine Learning IV (DML-IV) with Mundlak adjustments and Lasso-IV—compared to traditional 2SLS in panel data settings where the set of valid instruments is unknown. Using Monte Carlo simulations with balanced panel data (n = 5000, T = 5) and varying the number of potential instruments (p = 5, 10, 50, 100), it was simulated data-generating processes where only a subset of instruments is truly relevant for treatment assignment. Each estimator is assessed based on bias and root mean squared error (RMSE) over 100 replications to ensure stable performance metrics. Results shows that 2SLS performs well when the true strong instruments are known and included, recovering unbiased treatment effects with low RMSE. But it is not applied when instrument relevance is unknown or when instrument selection is required. Lasso-IV, which uses L1-penalized regression to select relevant instruments, exhibits upward bias and increasing RMSE as p increases, reflecting challenges associated with over-selecting weak or invalid instruments in high dimensions. In DML-IV with Mundlak adjustments achieves low bias and stable RMSE across all instrument dimensions, outperforming Lasso-IV consistently.

Keywords: 2SLS, Lasso-IV, DML-IV,Simulation, Panel data.

This research investigates the impact of foreign direct investment (FDI), government expenditure (GE), and inflation rate (IR) on the unemployment rate (UR) in Nigeria through the application of an Autoregressive Distributed Lag (ARDL) model. The study encompasses four ARDL models that analysed the effect of FDI on UR, GE on UR, IR on UR, and the combination of FDI, GE, and IR on UR over the period 1985-2021. Initial assessments involve a rolling correlation to test the significance of signals between each predictor variable and UR. Subsequently, the research employs the ARDL bounds test methodology to examine cointegration among FDI and UR, GE and UR, and IR and UR. Additionally, an Error Correction Model (ECM) is utilized to explore the causal relationship between these economic variables. The Augmented Dickey Fuller unit root test suggests that the variables attain stationarity at first differences (I(1)). Also, in determining the stability of the estimated parameters, the cumulative sum (CUSUM) of squares chart was applied. The results from the first model revealed that FDI at first lag significantly impacted on UR negatively at 5% in the long run but at current value of FDI the impact is insignificantly negative in the short run. However, the second and third models results revealed a positive insignificant impact of GE and IR, respectively, on UR at 5%. The fourth model results revealed that FDI at lag one negatively and positively influences UR in the long run and short run, respectively. However, GE and IR had positive but insignificant impact on UR at 5%. The CUSUM of square charts for the four models indicated an unstable parameter estimates during the sampled period. The study concludes that UR is not influenced by GE and IR in both the long run and short run but FDI influences UR positively and negatively both in the long run and short run, respectively.

KEYWORDS: Macroeconomic variables; Autoregressive distributed lag; Bounds test; Co-integration; Error correction model.

This study explores stress-strength reliability, R = P(X > Y ), focusing on the Akash distribution, which is well-suited for positively skewed data. We develop a Bayesian estimation framework using the Metropolis-Hastings (M-H) algorithm to estimate reliability. Building on Varghese and Chacko’s (2022) use of Maximum Likelihood Estimation (MLE), the M-H algorithm employs gamma priors for the strength and stress parameters and samples from their posterior distributions. Comparative analysis using jute fiber breaking strength data at 10 mm and 20 mm gauge lengths demonstrates that while MLE yields moderate reliability estimates (R = 0.5268) and good goodness-of-fit, the M-H algorithm provides substantially higher reliability values (R = 0.9767) with better robustness. These results emphasize the advantages of Markov Chain Monte Carlo techniques in reliability analysis, particularly in handling complex distributions.

KEYWORDS: Reliability, Maximum Likelihood Estimation. Parameters, Distribution, Metropolis-Hastings Algorithm.

This paper aims to demonstrate how new copulas can be derived through mixtures of copulas and their associated marginal distributions. We investigate the properties of these copulas by applying various dependency measures and conducting numerical evaluations via Monte Carlo integration. The study considers six bivariate copula families and asymmetric marginal distributions. To illustrate the methodology, we provide a simple example using the Clayton copula with exponential marginals, as well as the product copula with gamma marginals. In the applications, one of the six copulas is combined with the product copula, where the mixture is performed on the scale of the asymmetric normal distribution with an inverse gamma distribution. This results in an asymmetric Student-t distribution, generating a new copula. We estimate the model parameters using a Bayesian approach, with model selection based on the deviance information criterion (DIC) and similar metrics. For realworld applications, we apply the models to a dataset of insurance claims and two sets of daily returns from European and Latin American markets.

KEYWORDS: Dependency Measure, Markov Chain Monte Carlo, Model Selection, Scale Mixture.

Depression has now been known to affect the most people in the world. It is also known that students from low-and-middle income countries (LMICs) were also more likely to be predisposed to depression, with most studies reporting the prevalence to be more than half. A short 21-item BDASS-21 was used for primary data collection from the students of Gopalganj Science and Technology University, Gopalganj, Bangladesh. Descriptive statistics were used to explore the variations of the data. Chi-square test of independent and two-sample t-test was employed to test the association of categorical and continuous independent variables, respectively, with depression. Finally, a logistic regression model was fitted to check the influence of each independent variable on depression. Out of the 384 university students, 75.26 % reported depression. In the final binary logistic regression model, students from joint family had massive odds of depression (OR = 3.42, 95% CI: 1.61-7.27, p < 0.01) when compared to students from nuclear families. Similarly, the students that spent
more than 5 hours on screen had significantly greater odds (OR= 6.69, 95% CI: 2.81-15.96, p < 0.01) that those that did not. Another important significant variable was monthly family income, with students from the upper-class category less likely to report depression (OR = 0.35, 95% CI: 0.14-0.93, p = 0.04) than those in the Lower Class. The other significant variables were faculty of respondent, mother’s education, and father’s education. Three in four university students reported depression in our study. Students from joint families who also had a habit of using the internet more than five hours daily were in an increased risk group. Proper mental health support and awareness along with providing necessary resources to these vulnerable groups could help alleviate the current state of depression.

KEYWORDS: Depression, BDASS-21, Mental Health, Cross-Sectional Study.

This paper analyses the static and dynamic causal relationship between stock market performance and economic growth in India, along with other macroeconomic variables inflation rate and interest rate. The daily data on market capitalisation, Sensex, nifty50 and the value of shares traded are used as the measures of the stock market performance over the period from January 2014 to December 2024. In the empirical analysis, the ADF, correlogram, cointegration and causality tests are performed. The VECM is employed to analyse the causality between the variables modelling each of the variables individually as a function of the lagged values of all the variables. The estimated VECM results show that the dynamic processes converge as the estimated value of the error correction terms is negative but statistically insignificant. The study indicates that there is no strong long-run causal relationship between stock market performance and economic growth in India. 

Keywords: Stock market performance, economic growth, dynamic causality, VECM estimation.