- CFA Exams
- CFA Level I Exam
- Study Session 2. Quantitative Methods (1)
- Reading 4. Introduction to Linear Regression
- Subject 8. Limitations of regression analysis

###
**CFA Practice Question**

Which of the following statements is (are) false with respect to the limitations of regression analysis?

II. Regression equations cannot be used where the dependent variable exhibits exponential growth.

III. As the underlying relationship between the dependent and independent variables changes, a regression produced based on historical observations may no longer be valid.

IV. Widespread exploitation of any relationships which may exist between variables can quickly dissipate any benefits that may be had from its discovery.

I. If any of the regression assumptions are violated, then the results of the regression will be meaningless.

II. Regression equations cannot be used where the dependent variable exhibits exponential growth.

III. As the underlying relationship between the dependent and independent variables changes, a regression produced based on historical observations may no longer be valid.

IV. Widespread exploitation of any relationships which may exist between variables can quickly dissipate any benefits that may be had from its discovery.

A. I and III

B. I and II

C. II only

**Explanation:**While it is true that a regression is only valid for linear relationships, it is quite possible to transform the data so that the transformed data exhibits a linear relation. For instance, while sales figures may grow at an exponential rate, the log of sales will increase at a linear rate, equal to the rate of growth.

###
**User Contributed Comments**
3

User |
Comment |
---|---|

saaythong |
some regressions can either be relaxed or ignored and still have valid results (for example, some multicollinearity could still exist and have fairly valid results nonetheless) |

ThePessimist |
Yes, A is wrong. Notably, the independent variable could be non-random and still provide a useful regression. |

dblueroom |
When some of the assumptions are violated, such as no heteroskedasticity and no serial correlation, the estimated parameters are still valid, but we need to adjust standard errors of the parameters. So it is not completely useless. |