PLS-SEM with Time Series Data

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
Onur
PLS Junior User
Posts: 2
Joined: Mon Oct 29, 2018 8:05 am
Real name and title: Onur Tekel PhD(c) in Finance

PLS-SEM with Time Series Data

Post by Onur »

Dear all,

I am a PhD(c) in finance and nowadays writing my thesis. My question is whether it is possible to use PLS-SEM to deal with time series data. When I had a search in the web, I was not able to find that many studies performing such an analysis on time-series data. Thus, I require your assistance, if I may.

My topic is "Herding in Bank Loans" and basically examining the effects of economic conditions, banking industry conditions, regulation and credit risk on herding in bank loans (herding is basically banks' tendency to focus on a certain loan type collectively). I first thought that PLS-SEM is good for me, because a had a small sample (54 quarterly observations - 2004Q2:2017Q3) and no need normality assumptions. But then I worried whether I am in the right path, because there are not many examples in this area. Thus, I need to decide whether to continue with PLS-SEM or with another method.

Can you please provide your experince on this topic?

Thanks in advance,

Onur
jmbecker
SmartPLS Developer
Posts: 1282
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: PLS-SEM with Time Series Data

Post by jmbecker »

I think there are already some answers in this forum.

But briefly: You can model time series data with PLS but the quality depends on how your model would look like and whether you are able to take care of certain aspects.
There are currently no standard procedures available in the academic literature (and software) for PLS-SEM that would account for intercept heterogeneity such as fixed or random-effects panel models in regression. So basically using longitudinal data in a long-format where you have multiple observations for different times from the same subject would not be optimal in PLS. Beside the heterogeneity in the intercept (which would not be addressed) you also would need to adapt the bootstrapping to account for the clustering of the observations.
In contrast, using longitudinal data in a wide-format where each subject is a single row and you have the multiple times as different variables from which you create different constructs for different times and model lagged and cross-lagged effects should work fine.
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Onur
PLS Junior User
Posts: 2
Joined: Mon Oct 29, 2018 8:05 am
Real name and title: Onur Tekel PhD(c) in Finance

Re: PLS-SEM with Time Series Data

Post by Onur »

Tahnk you for your response Dr. Becker,

I think my study is one of those that may be classified as "not optimal for PLS". I attach an image of one of my models below. Basically, I have six latent variables namely economic conditions, banking conditions, credit risk, regulation strictness and investor/business sentiment all pointing at herding intensity construct (herding intensity: banks' tendency to focus on a loan type collectively).

Image
Post Reply