Racing To The Results Section

WELL,

Today's been a really long day. I ran a crazy long race with my friends, and it was pretty actually really exhausting. The problem was that I only had like 5 glasses of water before the race, and I ended up cramping a bit farther than halfway through. AT LEAST I FINISHED THOUGH.

Anyways, I'm going to start my blog post and talk for a bit because I really want to keep my mind off of what's going to happen to my body tomorrow.

So far I've already finished an outline of my results section, and I'm working on writing my results section up currently.

For the most part, among all finance studies that handle comparable firm multiples, the results section is rather short. Usually this is because researchers (such as Guo, Nissim, and Kim) simply present the general accuracy of the multiples and then move on to the discussion section to talk about whichever multiple performed the best and the reasons for its performance.

Because I am writing to a more lay, and less technical audience, however, my results section will greatly expand upon the format of the results section for most finance papers.

First, I will present the findings for the comparable firms of each of the five firms I analyzed within my section. The information will be in a table and will have each column as a different firm and its comparable firms. Thus, the table allows someone looking at the firms to determine which are more alike and which are more different. For example, Facebook and LinkedIn have a common comparable firm.

Next, I will establish the threshold for an effective multiple. Although this will only take a sentence, it is essential for later parts of the results section, which will use this threshold to draw conclusions from the data.

After establishing a threshold, I will begin presenting all of the data I have collected for the five firms. This will consist of five tables each with the harmonic mean, mean, median, and average accuracy of each of the multiples.

After establishing the table, I will then begin to make the first of my conclusions from the data in my results section. To determine which of the three methods (harmonic mean, mean, or median) is the best, I will present a table of p-values for paired two-tailed significance tests for the estimates. I was not able to conduct independent, multi sample significance tests, as the methods were not entirely independent of one another and relied on a common set of data that varied between successive trials.

Because my table shows that there is not significant difference between any of the three methods, to determine which multiple is the best, I will present a two figures. The first figure is to determine whether any multiple is significantly better than the others. I used the average error for each multiple (because the three methods were essentially the same), and performed significance tests two determine if the errors were different than one another. This would indicate that the multiple that caused these errors was significantly different from the others. Since each firm was a different "trial", the different trials are not independent of one another. Thus, I had to perform a paired two-tailed t-test for these values. My results show that no multiple is significantly different from the others.

The final figure in my study shows a comparative bar chart of the overall error of each of the multiples (taken using an average of averages) against the maximum threshold for error (15% error). The chart clearly shows how no multiple is even close to being considered effective. Thus, the final two figures helps me reach my ultimate conclusion in my results section: that no multiple is more effective than the others, and no multiple is effective at valuing Social Media IPOs.

Anywhoozies, that's all I have for now. I'm going to go sleep. (657)

Night,
Akash

Time to ANALYZE

HEY EVERYONE!

Data collection is finally over, and even though I have a lot more stuff to do, it's really nice knowing that I have everything that I need for my ultimate conclusions. That being said, the numbers that I've obtained aren't the best for what my original question was. The percent errors for almost all of my multiple estimates are close to 100, and only a handful are below 30 percent.

As mentioned before, this likely indicative of a more general characteristic of social media firms, which I plan to develop for my discussion section.

For now, I'm planning on developing exactly what I plan to do to analyze the data that I have collected.

First, I will compute the average error of the multiples for each firm. I have already done this, and plan on including this in a table for my data section. Thus, my data section will have five different tables with four columns and five rows. Each column for a single row will be the estimation error for a specific multiple and a specific firm. This will allow me to most directly compare the effectiveness of multiples for a given firm.

Next, I will need to compare the accuracy of multiples across the five firms I am analyzing. To do this, I will have a table for each multiple with the average error of each method for every firm. This allows me to determine which multiples perform the best for which firms, allowing me to draw conclusions for why specific multiples work better for specific firms.

Finally, I will determine which multiple is best overall across all five firms. To do this I will draw five box plots for each multiple representing the different errors for all five firms. I can then analyze the spreads for the box plots, and their average accuracy to compare how accurate they are in general, and how much they differ in their accuracy. The box plots also give a visual representation of the pricing errors, which allows me readers to more quickly grasp the differences between the multiples. To further explain the box plots, I intend to include a brief description of the box plots below them, summarizing important features of the plot such as the mean error of the box plot, which represents the mean error of the multiples.

I may also do the same box plots for each of the three methods that I used so that I can provide better comparative analysis of the data.

I think what I have right now is a pretty good start, but it might be hard to graphically fit five box plots side by side to ultimately compare them. If anyone has suggestions on how I can more nicely compare the data for the methods, it would be really useful. Anyways, thats it! (476)

Signing off,
Akash

The Search for Multiples

Hey All,

The past week has been amazing! For the first time ever, I don't actually need to go to school, and now I'm spending so much more time doing things I want to do, including research. I've picked up a Machine Learning Class from Stanford, and I'm also taking a Real Analysis Class from MIT OpenCourseWare.

Aside from all of this, I've made considerable progress with my research itself. At this point, I've written a program in C++ to sort the firms into comparable groups for the five firms that I intend on analyzing, and I've run the program on these five firms. The program took a bit longer than I intended to take (about a day and a half), but i'm still right on track for the data collection for next Sunday. At this point, I'm beginning to analyze the multiples for each firm within my question. Currently, I am working on the price-earnings multiple, and am calculating the error percentages for the different firms. I expect to be done with this multiple by tomorrow and begin working on the Book Value multiple right after that.

At this rate, I will easily have all the data that I need for next Sunday. Because the data needs to be in an organized manner, however, I'm just going to spend the rest of my post describing how I intend to organize the data.

For first picking the comparable firms for Facebook, Twitter, LinkedIn, Yelp, and Facebook, I intend to display the average values of total assets and EBITDA over the nine groups that I calculated for each firm, highlighting the one group each of the five above firms fell into.

Next, for each firm, I will have a excel spreadsheet that shows the pricing errors for each prediction using the different multiples. The rows will be represented as a different multiple, and the columns will consist of the comparable firm group, and the different mean, median, and harmonic mean values for their multiples. Finally, the pricing error for each of the above methods will be listed, allowing for a direct comparison of them.

In general, I'm feeling really positive about the entire process, and I can't wait to comment about my whole results next week! I'm just sad that most of the data collection part is coming to a close.

Signing off,
Akash

Numbers, Numbers, and More Numbers...

HEY EVERYONE!

We just got out of school, and the Superbowl just happened! Honestly, I'm not really much of a sport fan at all, but the food is still pretty darn good, and it's fun to watch people get really overly passionate and argue with each other. My research is going great right now, and its really starting to develop into something that I like doing.

Right now, I've just finished all of my data collection, and am in the process of finding the specific comparable firms for each firm that I outlined within my question. As of now, I've only found comparable firms for Facebook (which are Google, Apple, and Microsoft), but I have all the data for the other firms, so I don't envision my work being too much in respect to that.  I can get that done by Thursday, and that gives me a week and a half to begin my multiple analysis, which should be ample time given that it only involves comparing only a few numbers (only the range of the 50's, rather than the hundreds I've had to deal with up to now.) The big problem that I now am seeing for my firms is in respect to the comparable firm group itself. This problem really only surfaces with Facebook, but it may possibly have a big impact on the results of my study.

Essentially, the problem is that there are incredibly few firms like Facebook that are wildly successful, or at the "mega-capitalization" size. This essentially means that the number of comparable firms that I have to choose for Facebook is dramatically decreased. In fact, Google, Apple, and Microsoft are the only eligible firms at Facebook's level. I could potentially consider firms with fewer assets than Facebook, but this difference starts getting big very quickly: in nearly the 200 Billion Dollar range. I think I have found the very first limitation of my paper: that companies which are wildly successful are very difficult to have prices that are easy to predict because of the few comparable firms that they have. Although none of my other firms have assets as large as Facebook, and have comparable groups in the hundreds, I fear that this data may not ultimately be applicable to Snapchat. Snapchat's IPO has been valued in the 25 billion dollar range, something which is already incredibly high. Ultimately, I think I'll have to wait and see how the comparable firms for he other companies turn out, but I'm not too hopeful for the future.

This result could point to something more intrinsic to the social media industry for the future, however. Perhaps, because of these large social media firms already attracting consumers, there is such a large barrier to entry to the industry that only uniquely special, large firms that are difficult to value can enter the industry.

Either way, it seems like I have a handful ahead of me when analyzing my results. (495)

Signing off,
Akash