What's the limitation of regression model?

Can you send me the answer by email.

This is my economics regression model which going to predict AUD/USD as shown below: %AUD/USD= intercept+A1(%change in index of commodity price)+A2(%change in Australian current account)+ error R square is 0.49 What's the limitation of regression model? Regression analysis of the form assumes certain facts: Linear relationship between dependent and the independent variables. This means that the results are sensitive to outliers. Multivariate normality of the variables involved in the regression Absence of multicollinearity between independent variables, which means that the explanatory variables must not be correlated with each other. No auto-correlation, which means that the dependent variables across periods must not be correlated. For example if we look at stock prices, then price at day 't' is typically associated with prices on the previous day-'t-1'. Homoscedasticity The equation we are using needs to be checked for each of these assumptions. For example we are imposing a linear relation between variables, which may not be true, as borne by the low R2 value of .49. only 49% of the variation in %AUD/ USD data is explained by the two chosen variables- % change in prices and % change in Australian current account. It is possible that a cubic/ quadratic relation provides a better explanation. This can be tested by estimating a non linear regression equation that uses only 1 explanatory variable, like % AUD/USD= intercept + A1* % change in Australian current account +A2*(% change in Australian current account )2 + error. This can also take the form % AUD/USD= intercept + A1* % change in commodity price index +A2*( % change in commodity price index) 2 + error. The choice depends on which variable - % change in Australian current account or % change in commodity price index better explains % AUD/USD. From the nature of the data it seems that index of commodity prices and current account balances will be correlated, leading to multicollinearity. This is because current account looks at trade in goods and services with trading nations. This trade is expected to be determined by prices of these goods and services which is captured by price index to some extent. If the traded goods are part of price index then the two variables will be negatively correlated. As prices rise trade must fall. This correlation is called multicollinearity which casts doubts on hypothesis testing for any regression model. The t ratios of regression coefficients are low so that most null hypothesis are accepted- implying that coefficients are not significant, but the value of R2 is high. This is a contradiction as insignificant individual regression coefficients cannot lead to high explanatory power of the full regression model. We can also expect high degree of autocollinearity in this model. This is because the successive values of dependent variable- % AUD/USD are expected to be associated/ correlated. At any point / period the value of % AUD/USD will be associated with the previous days value of % AUD/USD. This is also common in price in stock markets as seen in empirical studies. Thus we can say that linear regression model has its limitations that must be checked in terms of the assumptions of the model before deriving any significance of the results