Bracket Buster: Part 2

In this post, we continue our analysis, and achieve better success after some adjustments to the methodology.

Our best indicator was tuition (in-state) which had 66% correct which beat out the analysts who performed the worst by 3% (1 Pick) and tied the analyst who picked second from last.

Our indicators didn’t stack up very well to the “experts” or national average at all. However, we noticed that our indicators were picking the wrong schools in completely obvious scenarios such as 1 vs. 16 matchups and other very high vs. very low seeded teams. These are generally the easy picks and not the ones we need help with. To counteract this we picked the top seeded team to automatically win in the following games: 1 vs. 16, 2 vs. 15, 3 vs. 14, 4 vs. 13, and 5 vs. 1. The top-seeded team was picked even if we know the higher seeded team won the game. Of these games there were four wrong picks out of 20 games.

We left 12 games to be decided by our indicators. These twelve games include the tough 8 vs. 9 matchup through the 6 vs. 11.

Figure 1 shows how these modified scenarios against the same unmodified scenarios.

Figure 1

As shown in the graph, the modified strategy beat the original strategy for every indicator. Every indicator went to 69% or greater and three strategies (Tuition (In-state), Admission Rate and 2012 endowment) either tied or beat the national average picks. Admission rates gave the best results at 78%.

Table 2 shows the performance with the comparators.

Table 2

The three top analysts and President Obama remained at the top, all picked over 80%. However, admission rate and 2012 endowment joined the upper ranks predicting better than the average pundits, the national average bracket and the “chalk” bracket.

By selecting the top ranked teams in the easy games and one of these indicators to predict the closer games our bracket beat the national average and the average analyst prediction.

Bracket Buster: Part 1

Round of 64: who predicts winners better?

Warren Buffett’s challenge of filling out a perfect bracket captivated America, but why? Who wouldn’t take a shot at winning $1 billion even if the odds are beyond farfetched? Historically, the highest payout for the lottery in the United States was $656 million in 2012 by the mega millions. Buffett’s prize is significantly higher and $1 billion certainly carries some shock factor. However, Buffets’ bracket challenge only lasted into the 25th game before the entire field of brackets were eliminated. After Memphis’s win over George Washington, all of the brackets were officially “busted”.

Did anyone really have a chance at winning the $1 billion bracket? There is a perception that choosing a bracket is less random than the lottery in most people’s eyes, but is it? Let’s put this is in perspective. The odds of winning the mega millions are 1 in 258,890,850 or roughly 1 in 258 million. The odds of picking a perfect bracket, assuming each team has an equal shot to win each game is 1 in 9,223,372,036,854,775,808. That’s 1 in 9 quintillion. The chances of winning the lottery are about 35.6 billion times greater. The lottery however has no skill or smarts involved.

If choosing the perfect bracket is not random, what criteria are used to predict winning teams? Some people choose based off of gut feelings, others are based on head to head matchups, some teams are chosen out of school loyalty, or teams that have been strong historically. But what if we looked at statistics having to do with the schools as predictors?

We looked at six different indicators relating to the colleges and universities participating in the tournament. These indicators include: admission rate, 2012 endowment, tuition (both in and out of state), and graduation rate (at both 4 and 6 years).

The indicators were chosen for various reasons. Endowment and tuition reflect the amount of money received by the school which helps fund sports programs and draw in potential players. Admission and graduation rates are also important factors in deciding on a school for student athletes. Very few college basketball players make it into the NBA and prospective players want to have their degrees mean something after basketball ends. We believe that better schools often draw in better quality players.

We used the same first round matchups as the original bracket and compared the real results to using one of the indicators as picking the winner. We only looked at the first round of the tournament.

Table 1 compares each of our indicators with the national average bracket, the top and bottom three performing analysts in the country, the president and the “chalk” bracket which always picks the higher seed.

Table 1

As seen in the table, our indicators didn’t fare all that well. The top three analysts and President Obama scored over 80% while our best indicator only scored 66%. The national bracket, or the most popular picks of each bracket in America (via ESPN) had 75% correct as did the “chalk” bracket which simply takes the highest seed.

In our next post, we’ll continue our analysis, and achieve better success after some adjustments to the methodology.

Going the Way of Blockbuster: Part 2

In the second part to our examination into what kind of impacts blockbuster drugs have on large pharmaceutical and biotechnology companies, more support comes in figure 3 which shows the index of total revenue and stock price for the period.

Figure 3

Revenue for these companies continues to remain steady despite large drop offs in revenue from Lipitor, Plavix, Seroquel and Singulair. While revenues remained steady, stock prices soared. The patent cliff was and still is a major concern for these large companies, perhaps revenues remaining steady and not dropping contributed to more investor confidence and the built in cushion for the stock price was initially a bit of an overestimation and prices corrected up.

To assess the loss of patent on a company level we took a look at Pfizer and one of the best selling drugs of all time, Lipitor, we see a major decline in sales but an increase in stock price.

Figure 31

It seems that these drugs aren’t as big a part of the total portfolio as originally thought since they don’t seem to influence stock price on their descent as much as they drove the price higher in their ascent.

Table 2 shows what percentage of total revenue these drugs had of their respective companies during Q1 2011.

Table 2

With the exception of Bristol-Myers Squibb and Amgen where Plavix and Enbrel made up 33% and 22% of their revenue, respectively, most drugs only made up less than 20% of total revenue. Singulair only accounted for 9% of Merck’s total revenues and Remicade was a drop in the bucket at 5%.

Our main conclusion from the research seems to be that if a company has a solid pipeline they should not be too worried about a blockbuster losing market share to a generic as the market seems to have priced this in. Companies need to be reinvesting their profits when they have a block buster in order to prepare for the future.

Data Sources

Sales data was provided by IMS Health via Drugs.com. Company fundamental data was provided by Charles Schwab & Co. Historical stock prices are adjusted closing prices provided by Yahoo! Finance.

Going the Way of Blockbuster: Part 1

How much do blockbuster drugs mean to stock prices?

Blockbuster drugs (annual sales > $1B) are important to the future success of pharmaceutical and biotechnology companies. DiMasi and Grabowski (2007) estimated that the average pre-tax R&D cost for approved biopharmaceuticals is around 1.2 billion USD.1 Besides putting the company in the black and rewarding shareholders, the additional revenue helps fund R&D and the purchase of potential drug candidates. New drugs are needed to sustain growth and probability since drugs on market lose their patent and fall victim to a steep decline in revenue due to the introduction of generic competition.

When generics are introduced and revenue for the blockbuster falls, it is often thought that stock price should follow. However, the change in stock price may be a smooth process because the market may price in pipeline drugs, or the market may have taken into account the impending loss of patent exclusivity.

We decided to try and see what kind of impact blockbuster drugs have on large pharmaceutical and biotechnology companies. In order to do this we created a list of the top 12 drugs according to United States sales in the first quarter of 2011. Of the top 12, we excluded ABILIFY® (aripiprazole, Otsuka) and ACTOS® (pioglitazone, Takeda) due to these companies not being traded on US exchanges. Table 1 shows the drugs and their associated companies which were included in the analysis.

Table 1

In total we looked at 10 different blockbuster drugs from 8 separate companies (three drugs from AstraZeneca). To get an idea of what percentage of total sales these drugs make up of these companies we totaled up their combined total revenues per quarter and the combined total sales of each drug in the sample.

Figure 1

Surprisingly, the total sales of these drugs together hovered between $8B and $12B while the total revenue for these companies fluctuated between $80B and $100B. Late 2011 and early 2012 saw a bit of a patent cliff for the big players, Lipitor, Plavix, Advair, Singulair and Seroquel all lost their patents. Lipitor, Plavix, Seroquel and Singulair sales dropped out of our dataset due to highly diminished sales which reflects the loss in sales in late 2012 through Q4 2013. In order to counteract this drop out of sales we assumed that subsequent sales per quarter would be 20% of the sales in the last quarter before patent expiration. In an IMS presentation titled “Spending on Medicines”, they report that 80% of a brand’s prescriptions are replaced by generics within six months of the patent expiration. 2

These drugs from Q1 2011 to Q4 2013 had a drop in total sales of 33%. With a 33% drop in drug prices total revenue for these companies only dropped 4% over the same quarter. The companies seem to be investing their profits wisely in drugs that can replace the earnings lost from the loss of patents.

In order to see how investors viewed the decline in sales of these block buster drugs we created an index for the stock price and total sales starting in Q1 2011 as our base. We then took the average of the sample and plotted them in figure 2.

Figure 2

The graph illustrates the divergence of stock price and sales of these drugs. This could be for several reasons. For one, as these drugs come off patent they are replaced by other drugs making close to, if not more in revenue. The other reason seems to be that the life cycle of these drugs is inherently built into the stock price. As long as the company has a solid pipeline and prospects the street doesn’t seem to worry about the disappearing revenue.

We believe both of these reasons have a high contribution to why stock prices keep increasing.

In the second part of this post, we will explore more reasons why the stock prices keep increasing, and conclusions from our examination.

References:

1 DiMasi JD, Grabowski HG: The cost of biopharmaceutical R&D: Is biotech different? Manag Dec Econ 28: 469-479 2007

2 IMS Institute for Healthcare Informatics.The Use of Medicines in the United States: Review of 2010.April 2011

Counting Medals: Part 2

Looking Closer At the 2010 Winter Games

Next, we did some additional analysis on the 2010 Winter Games by examining only countries that received a medal. In total this was 26 countries, the United States (37) received the most medals, Germany (30) placed second and the host nation, Canada (26) placed third.

We wanted to see if the specified outcomes of life expectancy, health expenditures per capita and GDP per capita had any correlation with the total number of medals by each country.

First, we graphed the number of medals by the average life expectancy of each country as shown in figure 1.

Figure 1

4

Alliance Life Sciences Consulting Group

There seems to be a small amount of correlation (0.244) between the number of medals and life expectancy. There are of course quite a few outliers as well. Russia came in with 15 medals, which was the 6th largest medal total, but had the second shortest life expectancy.

Most of the countries with high life expectancies saw at least a few medals except for Great Britain, which only received one medal and had one of the higher life expectancies.

The next outcome, health expenditures per capita had the highest correlation out of any of the outcomes (0.59). Figure 2 plots medals vs. health expenditures per capita in current US$.

Figure 2

The United States spend the most on health care per capita and also received the most medals. The other top spenders were Norway and Switzerland. Norway had the fourth most medals, but Switzerland only won the 11th most medals. Germany had the second most medals and spent the 9th most per capita on health care.

China, Russia and South Korea received great results with some of the smallest spending on health care per capita.

The last economic indicator examined was GDP per capita in current US$.

Figure 3

GDP per capita also saw a noticeable correlation with the number of medals won (0.45). Norway had the greatest GDP per capita and won the fourth most medals. Most of the top medal winning countries had a high GDP per capita. Most of the lower GDP per capita countries won less than 5 medals with a few exceptions such as Great Britain, Australia, Finland and Japan.

Overall, the highest correlations were GDP per capita and health expenditures per capita. Correlation does not always mean causation and in this case health expenditures per capita may not be as directly related to medal winning as it seems.

GDP per capita however may be much of the cause for winning medals. Higher income countries are able to afford the best equipment to train with, the best trainers and overall just another level of care for their athletes. Does this suggest that medals can in fact be bought?

Counting Medals: Part 1

Examining the relationship between health, economic outcomes, and medals at the Olympics

The Olympics are an amazing gathering of nations of different socioeconomic backgrounds from throughout the world. They all compete to bring home as many medals as possible. While the literal playing field may be level, figuratively it is not. Countries with more money can afford better training equipment, better trainers, dieticians and just about every other luxury imaginable for its athletes. Countries with more money also tend to be able to send more athletes to the games which increases the odds of obtaining a medal.

We decided to take the results from the 2010 Winter games (Vancouver, Canada) and the 2012 Summer games (London, England) and compare them to health and economic outcomes. We used indicators from the World Bank such as population, life expectancy, health expenditures per capita (Current US $) and GDP per capita (current US $).

Table 1 compares the average of these indicators from the countries that received at least one medal and countries that did not receive a medal.

Table 1

On average, for the 2010 Winter Olympics, the number of participants that countries with a medal sent was 89 as opposed to the 6 participants for thecountries which did not receive a medal. The summer Olympics followed the same trend with 114 average participants for medal winning countries and 9 for non-medal winning countries.

In both the 2012 and 2010 games the population of the countries with at least one medal was significantly greater than those without a medal. The life expectancy was also larger for countries winning medals as opposed to those not winning a medal. Notably in the 2012 Summer Olympics, countries that medaled had an average life expectancy of 74.28 years versus 66.83 for countries without medals.

Nations that won a medal also spent significantly more on health per capita than those that did not win a medal. The GDP per capita followed the same trend as above: GDP per capita was twice as large for medal winners in the 2010 games and 2.8 times larger for the 2012 games.

In our next post, with the 2014 Winter Games underway, we will do some additional analysis on the 2010 Winter Games by examining only countries that received a medal.

Hot Topic: Part 2

We wanted to see which study types and sectors these studies are being utilized. We went through title by title to mark which studies were of each type.

In total 12% of all studies had to do with cost-effectiveness, 2.5% budget impact, 1.7% literature review and .6% burden of illness.

The number of cost-effectiveness studies seems normal as does the number of budget impact since budget impact models are usually held close to the chest by pharmaceutical companies. The number of burden of illness studies seems quite a bit low for ISPOR.

Table 3 shows the type of analysis grouped by study type.

Table 3

All four types of studies were most utilized by cost studies including literature reviews. Literature reviews were the most split up of the group covering a wider range of topics and not as centrally located s cost-effectiveness and budget impact studies.

As expected cost-effectiveness and budget impact studies were highly concentrated in cost studies. Burden of illness had half of its presentations in cost studies and the other half was in clinical outcomes studies and health care use & policy studies.

Finally we analyzed the same types of study analysis according to the sectors of health care.

Table 4 shows the type of analysis by sector.

Table 4

This table gives a very interesting look into trends within each sector. For cost-effectiveness the most studies were in cancer, infection and cardiovascular disorders. This makes sense since these are some of the sectors with the most expensive treatments, especially cancer.

Budget impact studies followed the trend of cost-effectiveness studies closely except in neurological disorders and systematic conditions which had a significantly higher distribution of budget impact posters.

The sample size of burden of illness studies is small so it is tough to make a conclusion based on the distribution. Systematic disorders, cancer, diabetes and cardiovascular disorders saw the main grouping of burden of illness studies.

Literature reviews were the most evenly distributed type of analysis. The top three sectors for literature reviews were systematic disorders, cardiovascular disorders and research on methods.

Hot Topic: Part 1

Looking into abstract trends at ISPOR’s European Congress

Attending an ISPOR meeting is always a very interesting time. There are so many things going on such as classes, plenary sessions, workshops and poster sessions among various other activities. It can of course be quite overwhelming and an information overload for some of us.

For us, one of our favorite times are the poster presentations. It is really interesting to walk through and see exactly what everyone has been working on and it gives a great view of new trends and new techniques. With over 1,600 posters it is hard to find trends by walking the rows of posters and reading the titles. Fortunately, ISPOR posts the names of the abstracts for scheduling purposes. After each ISPOR we enjoy looking through the titles to analyze more closely.

In this post we dive even deeper to find the trends of this year’s ISPOR European Congress. We are focusing on the poster presentations made in the five separate poster presentation sessions. In total there were 1,736 abstracts accepted to ISPOR for poster presentation, 57 of those were withdrawn leaving 1,679 posters that were presented.

The posters are grouped by sector in health such as cancer, neurological disorders, infections etc. and then grouped by type of study such as cost studies, patient reported outcomes, conceptual papers etc.

Table 1 shows the distribution of total posters by health care sector.

Table 12

Research on methods, health care use & policy studies and cancer were the top sectors within the posters by a large margin. The usual suspects such as diabetes, infections, and cardiovascular disorders also rounded out the top of the list. A bit surprisingly, gastrointestinal and urinary/kidney disorders came in at the bottom with < 3% of posters.

While we did notice a large number of posters on HCV, these presentations were split between gastrointestinal disorders and infections. Without the split HCV was one of the more popular sectors.

In table 2 we grouped by type of study.

Table 2

The studies by type were extremely top heavy with the top 4 types making up over 70%. It was a bit surprising to see risk sharing only having 5 posters or .3% of the total posters.

After looking at the types of studies and sectors we decided to focus on the titles of the abstracts. We took all 1,679 titles and created a word cloud. The size of the word in the cloud is dependent on its frequency of use. Figure 1 shows the word cloud based on all the titles.

Figure 1: Poster Title Word Cloud

Figure 1

Of course patients and treatment were the most frequently used words followed by analysis, cost-effectiveness, economics, health, disease and cancer. The most common types of study types we saw were cost-effectiveness, budget impact, burden of illness and literature review.

Stay tuned for part 2 of this post, where we’ll continue our analysis.

Declaration of WAR Part 2

In this post, we continue our analysis, bringing salaries into the mix and calculating ICERs (Incremental Cost-Effectiveness Ratios) which we often use in evaluating treatments against one another.

We used the 2013 season salaries posted by Newsday. Generally we take the incremental costs and divide them by the health measure of incremental QALYs (Quality Adjusted Life Years). Here we have the metric of WAR, which serves as our incremental effectiveness because it is the number of wins above a replacement player: a minor leaguer or bench player. To calculate incremental cost, we subtract the salary of this minor league or bench player; for this analysis we are using the MLB league minimum of $490,000. Our final equation looks like:

ICER

We began the analysis by looking at the most cost-effective players in the league projected for an entire season. These players were mostly paid right around league minimum so in order to reward the top players we only looked at those with a WAR > 3. This made the analysis more about being the better player than having a positive WAR and a salary as close to league minimum as possible.

table 3

As expected with a large gap in wages throughout the league and a smaller gap in WAR the most cost-effective players were those making very close to league minimum. Most all of these players are budding superstars on rookie contracts and will see large pay raises in the future to complement their on the field achievements.

On the other end of the spectrum we looked at players who had extremely high ICERs or were dominated by their replacements. We had over 200 players in the dataset that had a zero or a negative value for wins above replacement. Most of these players had less than 250 PAs projected so we filtered to include only players with 250 or more projected plate appearances. Vernon Wells and Brendan Ryan (both Yankees) had a projected WAR value of 0 for the price of $24.6MM and $3.25MM a year, respectively. Adeiny Hechavarria for $2.5MM had a projected WAR of -0.3.

We decided to look at which players that were projected to actually make a positive impact cost the most. To do this we only looked at positive WAR and players with 250 or more projected plate appearances.

table 4

A-Rod led the way. While still providing definite value to the Yankees, his salary could be considered too high given his production. A trend starts to develop here as the teams that house these players tend to be in the lower portion of the standings. Teams may allocate a lot of funds into one player who underachieves and they are left with smaller salaries for their other positions.

Finally we wanted to show the projected most cost-effective players at each position. These players would be an owners dream. They are all extremely inexpensive compared to older top players and are fantastic players. In order to make sure we were truly finding the most valuable players and not just the players with a positive WAR and had a salary closest to league minimum, we selected only players with a WAR over three for all positions except for DH which included all players with a WAR over one.

We also took into consideration the projected time each player would spend at each position and weighted the salary accordingly. The players WAR is unique to that position so salary was made to be unique for percentage of time spent at that position as well.

Table 5

It’s not a surprise that the most cost-effective players at each position in the league are all young, budding superstars. Teams that draft well are able to add significant talent without the large price tag. Wilin Rosario provides an excellent boost for the Rockies at catcher while Andrelton Simmons is the best valued shortstop, and overall player in our analysis.

Building a team is all about allocating your resources in the most efficient way given a certain budget constraint. In this modern day and age we are seeing an increase in the use of advanced statistics to evaluate players which is giving a boost to some of the smaller market teams. Hopefully these trends will continue which would make for better quality baseball all around!

Declaration of WAR Part 1

Using WAR and salary to calculate incremental cost-effectiveness ratio for MLB players

Baseball is a statistician’s dream. Unlike other sports such as football, basketball, and hockey which generally have a continuous flow with a clock, baseball is a game of many independent events. Each pitch presents a new opportunity and every player provides some type of unique value to a team.

More and more, we are seeing the use of advanced statistics from general managers and executives when assembling teams. This is done for large market teams as well as small market teams. The 2011 movie Moneyball chronicled (semi-accurately) how the 2002 Oakland As team was put together using sabermetrics. 1 They looked for players that could provide the most value to the team for the smallest salary.

Here, we aim to look at the players who provided the best, and the worst value to their teams. We utilized the Fan Graph‘s depth chart database which has projections of WAR over an entire season. We decided to focus on the wins above replacement (WAR) metric, which shows the number of additional wins a player provides over a bench or minor league player.

Before we bring cost into the mix we compiled a list of players with the highest and lowest projected WAR at each position excluding pitchers.

Table 1Table 1

As expected this list is an all-star caliber roster of superstars. Mike Trout led all players with a projected 7.8 wins above replacement out of center field. The AL and NL both own four of the spots on the list excluding DH since very few NL players have a positive WAR value in the DH position.

We also created a list of the worst projected players at each position which is shown in table 2.

Table 2

In order to qualify for the list of lowest WAR, players must have had a minimum of 250 plate appearances at their respective positions. We tried to single out the worst “everyday” players. The Blue Jays’ DH led the way with a projected 0.4 less wins than a replacement player would have. As expected the Marlins and Blue Jays dominate the list with 4 and 3 players, respectively.

We also decided to total up each position for every division to compare the strengths and weaknesses across divisions. Table 3 shows the sum of projected WAR for every player in each division.

Table 3

Overall the AL is projected to have more of a balance than the NL. In the NL, the West led 5 separate positions while the East was worst in 6 separate positions.

In our next post, we bring salaries into the mix and we calculate ICERs (Incremental Cost-Effectiveness Ratios) which we often use in evaluating treatments against one another.

1 In our opinion, the success of the A’s team can be attributed to the big three: Barry Zito, Tim Hudson, and Mark Mulder. These pitchers were not the focus of the Moneyball story.