Hello, I'm Professor Brian Bushee, welcome back. In this video, we're gonna take a look at models of discretionary expenditures. Managers can also manipulate their earnings by changing their cash expenditures on things like research and development, advertising, training costs, and so forth. That's because the cash spending for those kind of costs are expensed immediately which means if you delay some research and development spending from this year to next year you also shift the expense from this year to next year. There by making your earnings look better. We're gonna use the same approaches with discretionary accruals, where we're gonna use regression techniques to model the normal level of these expenditures, then look for differences from these normal levels as potential evidence of earnings management. Let's get started. So if you recall, when we talked about discretionary accruals, we divided net income into cash earnings plus non-cash earnings. And discretionary accruals were the way to manipulate non cash earnings. While discretionary expenditures are the way to manipulate cash earnings. So, managers can manipulate cash earnings by delaying or accelerating real expenditures. Especially those that have to be expensed immediately according to the rules. In other words, they can't be capitalized and amortized. So this would include things like research and development expense, advertising expense, and selling general, and administrative expense which includes things like maintenance on machines, employee training, employee travel. And for some companies also includes the R&D and advertising expense. This real earnings management does not violate any securities laws. Because managers are not playing with the rules, they're not playing with estimates, they're just changing the timing of when they spend cash. And there's no rule that says when you have to spend cash. And as such, it's viewed by managers as a much more ethical form of earnings management. The drawback is it has to be done well in advance of the end of the period. So, accruals, because you're playing with assumptions, or estimates, you can change those on the last day of the period. But for real earnings management you have to start delaying or accelerating cash payments well before you get to the end of the period, so there has to be some advance knowledge that your earnings are gonna be in trouble. Real Earnings Management is much harder for outsiders to detect because we always have the concern that managers may have a legitimate reason to delay expenditures due to their poor performance during the period. >> Yipee. I was really hoping we would talk about ethical earnings management. It is an open secret that my least favorite form of speech is an oxymoron. >> Yes. Ethical earnings management is an oxymoron, but there's been a lot of survey evidence that finds that managers view this kind of earnings management as not unethical. They find it as legitimate. And that's because they're going through the following logic in their head. It looks like our performance this period is gonna be below expectations, which means times are tough, we're not performing as well. We need to scale back on some of our cash costs so why don't we reduce or delay some of our RND or advertising? Or SG and A from this period into next period. And again that could be legitimate because if things are not going well at the company, maybe you do wanna cut some of these costs. But the timing of when these are done and the amount that these things are reduced all suggest that they're doing it really to try to make their earnings look better and meet an earnings target. So one of the things we'll work on in this video and next video is to try to disentangle legitimate cuts in these expenditures because the company is really performing poorly, from cuts that managers may say are legitimate, but really are intended to make their earnings look better. So now let's talk about the model that we're going to use to try to estimate discretionary expenditures. So the normal level of these expenditures should be a function of last year's expenditures, as well as the company's revenue and its growth in the business. So we're gonna model changes in expenditures. So we'll take the current year minus last year. So for research and development, we would take this year's R&D expense minus last year's R&D expense to recognize the fact that last year's expenditures have a big determination on what you're gonna spend this year. We're gonna use prior years' revenue growth for the growth in the business and then prior year's revenue. We wanna use the prior year because normal expenditures are often budgeted based on prior year's results. So after you see the results for one year then you budget what you're gonna spend for these expenditures the next year so that determines the normal level of expenditures. Also using the prior year insures that the model is not affected by current year sales manipulation so if we used revenue in the same year. Maybe managers are trying to manipulate sales as they try to manipulate these real expenditures. And just like the Discretion Accruals model everything will be deflated by prior total assets to take out the size effect. So the model is gonna be that the change in expenditures is gonna be a function of the prior sales growth and the prior sales. For expenditures we'll use either SG&A expense, R&D expense or advertising expense. We'll do separate models for all three and you'll want to look at all three because you don't know X and T which one the company might be manipulating. We're gonna run industry year regressions to get the estimated A B and C parameters, so we're not gonna fool around with the time series stuff with the old firm. We're just gonna do the regression within industry. Once we get the parameters then we can calculate the normal change in the expenditure as the intercept A, plus the coefficient B times prior sales growth, plus the coefficient C times prior sales. Then discretionary expenditures will be the total change in the expense minus the normal change in the expense based on this equation. And again we'll do it separately for SG&A, R&D, and Advertising. >> What a complexly simple model. I love the redundant originality of your use of both sales and change in sales. You are a vapid inspiration to us all. >> I see what you did there by putting in some more oxymorons in your question. Yes, this is a very simple model. Academic research has tried to do more complicated models, they've tried different models for different types of expenditures. But the added explanatory power in doing so was just not that great. So I decided to just keep the model simple in this case because they work pretty much as well as the more complicated models. So, instead of adding complexity for a tiny bit of extra explanatory power, I wanted to keep it simple, and these models will be good enough for our purposes. So let's go to the spreadsheet for Arfabark company where we'll look at an example of how to calculate the model. We'll again focus on 2009 as a potential year that they might have manipulated. I pulled in probably way more years of data than I need because we're not gonna do a time serious model, but at least you need some prior data to look for time trends. The first few columns are all the raw data that I need to calculate the model. The purple columns are the variables that go into the model. The orangeish columns are the intercept and coefficient parameters that are estimated from industry year regressions. That's why they change every year. And so I will also give you a spreadsheet where you can pull in these parameters for the industry and a given year. So then you can use these estimated parameters and the variables to come up with normal R&D. Take the total R&D minus the Normal R&D to come up with the Discretionary R&D and if we look in 2009 there's a small negative. 2010 there's a small positive but these are within the realm of, sort of, normal, bouncing around and discretionary R&D. So I think we would conclude from this that there's no evidence of manipulation using R&D in 2009 or 2010 which is not a surprise cuz we didn't really expect there to be strong incentives. So to see if the model works let's do a manipulation. So in 2009 I drastically cut R&D. And if we look at the discretionary R&D number it becomes much larger negative. So it's -0 .019. Now remember, if you want to manipulate earnings upward with a discretion expenditure, you need to cut it, right, cut the expense to make earning higher. So negative discretion RD would be indication of trying to make earnings go higher. And notice in the next year we get a big positive. That's because if you cut R&D in one year, and then don't manipulate in the next year. It will automatically swing up to a big positive because the model is based on this change in R&D. So it does look like the model picks up a cut in R&D in this case. >> Looks like a bluntly sharp model to me. Wouldn't the deceptively obvious huge decline in R&D suggest that there was a problem? >> Yes, in this simple example, the R&D cut is pretty huge and it should raise some suspicion. But keep in mind that a cut in R&D alone is not evidence of earnings management. We have the model because the cut in R&D could be expected based on change in business conditions. The model helps us find those situations so that we can find cut in R&D that is not explained by changes in business situations and is more likely to reflect manipulation. What we'll do in the next video is we'll bring in some quarterly changes in R&D as additional evidence to try to help us spot manipulations. But a cut in R&D alone is not enough evidence, because it could be expected based on changes in the business. In this spreadsheet I also did discretionary expenditure models for SG&A and I'll let you look through this on your own at your leisure. But you can see I put in a manipulation and we do see an increase in discretionary SG&A an increasing negative number which would indicate they cut SG&A to make their earnings higher. And then I do it for advertising. We see again the effect of a manipulation, advertising shows up in the discretionary advertising number, but in this case it's really small magnitude because advertising is not a huge expense for Arfabark. In general, you'd wanna use the discretionary expenditure model for the expenditure that's a big deal for the company. So if it's a big R&D company, you want to focus on R&D first. If it has huge advertising then it might make sense to focus on discretionary advertising but if they don't have a lot of advertising it's probably not the tool that's used for manipulation anyway. So hopefully that gives you a good sense on how to estimate these discretion expenditure models for R&D, advertising, selling general administrative expense. In the next video we're gonna do a couple refinements to these models, and then look at three cases to see how well they work in detecting earning management. I'll see you then! >> See you next video!
