Відео

Thanks a lot 😊

You're welcome

Yes, it refers to OLS regressions covered in previous videos

@@YUNIKARN I'm just learning about OLS regressions, as part of my Econometrics I course, and I have issues understanding what it means for X to be (or not) stochastic. I read on Wikipedia that it was called "weak exogeneity". After reading on Gujarati and Porter's book, I thought "if X can take many values, then it indeed could be stochastic". But then I wondered, "what could it mean for it not to be stochastic? Would it imply that there's a unique universal value for X? Like everyone having a single income, or every country having the same GDP?" And I couldn't really figure it out

@@tomasbeltran04050 I assume X refers to the independent variable (explanatory variable). Explanatory variables are observed (i.e., we observe a value say GDP growth = 2.5%). X is not stochastic as the value is known. There is no uncertainty (variance = 0). For instance, we might observe the values X_i = {2, 4, 6, 7} for four cross-section units (e.g., countries). These values are observed and not stochastic. You find quite a few videos on regressions on my channel (using Excel, Stata and Python). May the Force be with you!

I am glad you find the video useful!

Sure. y_{it} = \alpha_i + \beta x_{it} + \gamma z_{it} + \delta x_{it}z_{it}.

@@YUNIKARN in that case how would one go about interpreting the results

​@@uwuslayer0211 You estimate the coefficients using fixed effects. Then you determine the partial impacts of x_{it} or z_{it} by taking partial derivatives. This gives: dy_{it}/dx_{it} = \beta + \delta z_{it}. Hence, the effect of x on y changes with the value of z (moderation).

@@YUNIKARN The model that I'm running imgetting a very low r2 for within (around 0.15) while a very high (0.89) for between when I run the fixed effect regression in Stata. What could be the reason for this? Another thing I wanted to ask was , what is to be done if the coefficients of fixed effect lsdv and the within effects are different? From my understanding, they are supposed to be the same

@@uwuslayer0211 Your results are pretty normal. It is common to report the overall R-squared (combines within and between estimates). The LSDV (least square dummy variables model) is the fixed-effects estimator. You just add dummies to the OLS regression. They should match. Check whether some dummies drop

The datasets and updated codes / videos etc. are available on Udemy: www.udemy.com/course/getting-started-with-stata/?referralCode=337F796C4A4C63DD833F

Thanks, well coffee (no milk, no sugar) is fine - but not too much (affects sleep) ... just having coffee

Thanks for your comment. Yes, PyPDF2 merged into pypdf - and there is PyPDF3 and 4 that did not (yet) merge. A bit messy. Best to work with virtual environments and control the version. Thr PdfWriter() class in pypdf works similarly. I should give pikepdf a serious try ... the PDF fun does not seem to stop

Physics (thermodynamics) suggests that the weight change \Delta w_t is explained by the calory consumption c_{t-1} and the body's current maintenance level denoted m (which is assumed to remain stable over time - some variation can be introduced). The regression becomes \Delta w_t = \beta (c_{t-1} - m) + \epsilon_t. So assuming that \beta>0, higher calorie intake than maintenance increases body weight. Rearranging the equation gives: \Delta w_t =- m \beta + \beta c_{t-1} + \epsilon_t. Estimating the coefficients in this regression model provides a constant term (\alpha = - m \beta) and a slope coefficient \beta. By knowing \beta, one obtains maintenance m. Note that one could permit that m varies over time (time-varying coefficients). These can also be estimated using rolling regressions. So much fun!

All Stata codes are on GitHub. The link is in the description. Datasets are available on Udemy (Getting started with Stata - see link in the description).

A fixed effects model allows (only) a shift in the intercept but the slope coefficients are constant (fixed). Interaction effects go further by permitting changes in slope coefficients. They can be combined with fixed effects - but I would not recommend using xtreg, fe (the default) as you might like more control over the dummies used in fixed effects models. I tend to prefer xi: reg y x i.code and add the cross-sectional dummies and maybe additional shifts in slope coefficients.

@@YUNIKARN Thank you so much!

Thanks for your comment. The brackets affect the link. I adjusted the description github.com/GerhardKling/PythonForDataScience

You’re welcome 😊

Thanks, much appreciated!

Glad you enjoyed the video. Yes, Stata has the command kpss, which conducts the KPSS test. You might need to install it depending on your version of Stata

That sounds too good to be true. I will respond to your email by tonight. My Python code can be adjusted to test any other betting strategy. The link to the GitHub repository is in the description

I believe that the codes should be on GitHub github.com/GerhardKling/Research

Here is a link to my GitHub for this playlist: github.com/GerhardKling/FinancialModelling Of course, this playlist is for free. I will release one video each week. There is an updated and more comprehensive course on Udemy: www.udemy.com/course/company-valuation-a-guide-for-analysts-investors-and-ceos/?referralCode=FC67D2F6397D78F56448 I usually post coupons for discounts on my channel page - but I only get them once a month. I have spent a lot of time in India. I look forward to my next trip in December!

Welcome. Thanks for watching

Welcome! Enjoy the Joy of Data Analysis

😊

Glad you liked it!

Yes, I did wonder as well. I check his content - but it seems to be all removed. Let's hope he is fine

What exactly is confusing about it?

This is the way. Watch to the end!

In panel data, the standard approach is to use the cluster option in Stata, which also accounts for serial correlation.

You can use time and firm dummies. However, there is a risk of overfitting. You might get great in-sample model fit - but once you try to predict out-of-sample the quality drops.

Thanks, much appreciated!

It should be here: github.com/GerhardKling/DataScienceStata/tree/main/Lecture4

@@YUNIKARN thank you

If you are interested in pursuing a PhD at the University of Aberdeen, we can have a chat. Otherwise, I cannot provide any individual consultation due to regulations. All the best!

@@YUNIKARN its ok, let us talk for advice and make decision together , thank you

@@YUNIKARN I am still waiting for your support , let us talk and see if i can pursue it

I responded. If you want to study at the University of Aberdeen, I am happy to help. However, I cannot provide individual consultations

Glad it was helpful! May the Force be with you!

Well, the truth hilarious. I am working on a panel data course. You find two Stata courses on Udemy (www.udemy.com/user/gerhard-kling-2/ ). May the Force be with you!

I am glad the videos are useful. I am working on a panel data course (GMM, difference-in-difference etc.). Put video editing takes a very long time. Enjoy your research project!

Two years is short. Most PhD programmes will require three years. So you will need to think about self-funding one year or finding alternative arrangements (write-up periods etc.). I worked at Utrecht University. The Dutch system is quite different from the UK as most PhD students have a teaching position. Do you want to try the Dutch market or UK/US? You need a good fit with a potential supervisor. Good luck!

@@YUNIKARN It's entirely up to me, by the way; money is not a problem, nor is getting employment. I am prepared to provide for my own needs in order to maintain my research's flexibility. This is the reason I asked you to share your thoughts on the matter. I want to work closely with the university of my choice and be totally independent. That's the reason I'm funding myself. But I believe that two years is reasonable. I wish not to spend more time. I am at Erasmus University and TU Delft

@@alimtanta2934 I had PhD students who completed within 2 years (both with industry experience). With a good proposal, you should be able to get admission. I suggest contacting potential supervisors before making an application. Good luck!

​@@YUNIKARNThank you very much for your response. Your channel is very educational. I appreciated

Yes, I take it as a compliment

Glad it helped!

There is an implementation in Stata called xtbreak. My Udemy course on Applied Time Series explores structural breaks in a lot more detail (testing for knowns and unkown breaks)

@@YUNIKARN thanks for a good reply , now I have used the test and found two breaks in the data and wanted to use Lagrange multiplier Lee and Strazicich (2003) but could not find the command in stata I searched online could you please help with the command if it available in stata or even are thanks

@@bolisolwak1390 Only available in R or RATS as far as I know. Here is a link to the GitHub. So either run it in R/RATS or translate into Stata github.com/hannes101/LeeStrazicichUnitRoot

@@YUNIKARN Thanks once again, I am not familiar with that software if I could get code for R it would sound good to me do you have the link for R ?. Thanks.

@@bolisolwak1390 I suggest running R through Anaconda (www.anaconda.com/). Free download for R and R Studio. If you have not used R before, it will take a while to get used to

Thanks for watching! I am looking into the job market. It is not easy to get reliable data. On Monday, I will present some data on foreign direct investment. India or China - who is ahead?

A pleasure. Thanks for watching!

I am glad the video was useful. May the Force be with you!

Thanks, they did an awesome job!

Thanks, much appreciated. May the Force be with you!

Yes, N refers to time. Everything is discrete

You're most welcome

I have done more on panel data in my Udemy course - see www.udemy.com/course/getting-started-with-stata/ I am also working on new material but this will take quite a bit of time. May the Force be with you!

I did a video on "Why does Stata drop my variables?", which will be published on Monday 1st April 2024 - April Fool's Day Special. Link: ua-cam.com/video/CgHueBDaEnU/v-deo.html If you have four dependent variables, it suggests that you will have four separate models to explain them. Of course, you can have FE or RE as these models are separate. If you do not know which one is actually dependent, then you need to explore causality (or causal order if you have a time dimension). May the Force be with you!