This post is available as a PDF download here.
- Benchmarks can be a very difficult subject to pin down. Choosing different ones can create drastically different backdrops to frame both short and long-term results.
- This was even true in our 2018 March Madness Bracket Challenge, with the value-weighted benchmark taking the top place.
- As investors, we must constantly battle behavioral biases such as hindsight, anchoring, and confirmation.
- Managing these biases isn’t that important when competing in a one-off tournament like a bracket pool, but when we have constant opportunities to alter our portfolios based on current information, avoiding the compounding effect of our own behaviors can be one of the biggest determinants of long-term success.
For March Madness this year, we put on our own bracket challenge based on some of the work that Resolve Asset Management has done in previous years. Primarily, the competition aimed to:
- Encourage a larger sample size of teams to be picked rather than just straight chalk bets.
- Reduce the risk of legacy errors in brackets. Virginia is a great example of this this year.
We tweaked the rules a bit, and you can find them along with the subsequent recaps in the related posts.
- The rules (https://blog.thinknewfound.com/2018/03/newfounds-2018-march-madness-bracket-challenge/)
- Some tips (https://blog.thinknewfound.com/2018/03/march-madness-for-investors/)
- Sweet 16 Update (https://blog.thinknewfound.com/2018/03/newfound-2018-march-madness-sweet-16-update/)
- Final 4 Update (https://blog.thinknewfound.com/2018/03/newfound-2018-march-madness-final-four-update/)
There is not much to write home about from the final three games of the 2018 tournament. The results were pretty chalky with Michigan and Villanova making it to the final and Villanova taking the title.
However, now that the tournament is all wrapped up, we’d like to memorialize this year’s challenge with some of the lessons we learned during the madness.
First off, we would like to congratulate MovalTrend, our Newfound 2018 March Madness Bracket Challenge (non-benchmark) winner! You can rest well, knowing that you were the top team-picker.
Which Benchmark You Use Matters
Perhaps the Final Four Update spoiled the end result given that it was impossible for any of the brackets to overtake the seed-weighted (value) benchmark. That bracket did so well because of the history-making upset of 1 seeded Virginia by 16 seeded UMBC. This result was also a boon for the equally weighted bracket, which took second place.
However, there were three other benchmarks added to the bracket pool in the tournament. The inverse seed weighted (momentum) benchmark, the inverse points-per-win (PPW) benchmark, and the front runner benchmark did not fare nearly as well as these other two, taking 11th, 28th, and 45th place, respectively.
You will almost always be able to find some benchmark that beats your portfolio. Conversely, you will almost always be able to find a benchmark that your portfolio beats!
Putting thought into what is an appropriate benchmark is a worthwhile endeavor.
This is especially important when setting appropriate shorter-term expectations for investments.
Consider a simple trend following strategy on the broad equity market. While it is well documented that this strategy has outperformed over a full market cycle, the shorter-term performance can vary substantially.
If we compare the strategy to equities, it can experience large tracking error over one-year periods with most of the one-year returns failing to keep up with equities.
Rolling One-Year Outperformance of Trend Following vs. Equities
Data Source: Fama French data library. Calculations by Newfound Research. Past performance is not indicative of future returns All performance is hypothetical and backtested. Performance assumes the reinvestment of all distributions. Returns are gross of all fees, including management fees, transaction costs, and taxes.
However, if we benchmark it to a 60/40, we see a different picture: one with lower absolute tracking error and a positive median outperformance for the tactical strategy.
Rolling One-Year Outperformance of Trend Following vs. 60/40
Data Source: Fama French data library. Calculations by Newfound Research. Past performance is not indicative of future returns. All performance is hypothetical and backtested. Performance assumes the reinvestment of all distributions. Returns are gross of all fees, including management fees, transaction costs, and taxes.
In this year’s bracket challenge, yes, the value weighted benchmark won. But it could have played out very differently if the Virginia upset had not occurred. Perhaps the momentum portfolio would have done better in that scenario.
For this tournament, the risk-free inverse PPW is likely the most appropriate benchmark. This bracket earns the same point total regardless of the outcome.
Because the points awarded for each win were tied to each team’s expected number of wins, every bracket started off with the same expected return. How much tracking error you opted for is what caused the differences.
Through that lens, anyone who beat the risk-free portfolio showed that the risk that they took paid off.
We can treat the value and momentum benchmarks simply as a systematic way to allocate to teams, similar to how these factors are used to construct investment portfolios.
The Unexpected Can Have a Big Payoff … With a Lot of Risk!
Looking at how many points each team ultimately could have won shows us where the opportunities were.
Fraction of Points Awarded to Each Team
This is essentially an attribution of the equal weight portfolio. Half of the total points were won by Maryland-Baltimore County (16) and Marshall (13). Another quarter of the points were won by six of the 7 through 13 seed teams.
Concentrating in the lower seeded teams may seem like the obvious solution going forward since they paid out much more than higher seeded teams. But remember that all the 8, 12, 14 and 15 seed teams and the other three 16 seed teams all got zero points, as did many other sub 7 seed teams. Betting on 8 teams out of 40 yields only a 20% chance of landing in this 75% slice of the pie.
You needed to have some skill.
Let’s take a closer look at the MovalTrend’s bronze-winning portfolio.
We see larger allocations to all 11 seed teams with some 12s, 13s, and 14s thrown in.
From an attribution standpoint, Loyola (11), Marshall (13), and Syracuse (11) carried the bracket bringing in 85% of the total points.
Including allocations to some of the lower seeded teams that won was a necessity in this tournament, but the allocations did not need to be huge.
In fact, large allocations to individual teams were discouraged this year.
Behavioral Biases Played A Part
One of the rules in this year’s challenge was that concentrated portfolios would be penalized by up to 50% and diversified portfolios would receive up to a 25% bonus.
Despite this, if you randomly went all in on one team and just settled with getting hit with a 50% penalty, you still would have had a 4.7% chance of winning the tournament and a 9.4% chance of beating all the human-made brackets, assuming that all other brackets were unchanged.
If you restricted your field of teams to choose to the bottom half of the seeds, your odds would have improved to 9.4% for an overall win and 15.6% for a win, excluding the benchmarks.
While these percentages are still higher than what we – as the rule-makers – ideally would want with a tournament that is intended to isolate skill, it didn’t matter because everyone made sure that their bracket was diversified enough to avoid the penalty.
When there are known penalties and unknown payouts, it can bias our course of action. Subjecting your bracket to a 50% penalty was a big hit to take right off the bat. Getting a 25% bonus seemed like a good choice for everyone who participated.
Discussions like this frequently come up when talking about fees.
Paying 1% for a strategy that can be 100% tactical is more cost effective than paying 0.6% for a strategy that can only be tactical on 50% of the portfolio.
Self-implementation for the passive portion of the portfolio makes this more apparent.
If the goal with an equity portfolio is to be 80% strategic and 20% tactical, implementing it using an essentially free passive sleeve of ETFs paired with a sleeve of tactical managers with a fee of 1% is possibly a better option than paying someone even 30 bps to manage the entire portfolio.
Our own behavioral biases can often prevent us from getting to the portfolio that meets our objectives.
You Still Can’t Win With Hindsight…
In our commentary published before the tournament started, we discussed how hindsight is only beneficial if we learn from it going forward. We then discussed the risk of overfitting a model based on what we have observed historically and ending up with one that is not robust going forward.
So now let’s answer the question that is on all of your minds: would my entry have beaten the benchmark (and all of you)?
This year’s rules were similar enough to what we used in 2016 that I can run my methodology from that year with the 2018 data known before the tournament started. At a high level, my process was to minimize the risk of finishing worse than the top 10% of the field based on simulated outcomes.
The result was a very diversified portfolio with some allocation to every single team. The chart below shows the allocations over 2%.
Predictably, the largest allocations went to the teams most favored to win (and we know how that panned out). However, the small allocations on the less favored teams led to a decent amount of points for the upsets.
Alas, I still didn’t beat the seed-weighted or equal weight benchmark, although I would have finished in the top 5 (largely because of my UMBC allocation). I still would have had to bow to MovalTrend.
When my model’s objective was to finish in the top 10%, I can’t complain that it did exactly that.
This is yet another example of the trade-off between how well a quantitative model fits the training data (e.g. a backtest) and how robust it is on new data. The problem with making the objective to finish first in my model is that it usually fits to noise. Apparently, I will have to increase my skill for picking teams if I want to improve my rank.
Speaking of skill, one entry that was not submitted was Corey’s. How might he have fared?
I’ll leave you with this conversation we had on the day that entries were due:
Sometimes the right answer might be one of the simplest.
Tweaking the Rules
One thing I noticed in the results this year that I would like to address in a subsequent tournament iteration relates to sequence risk. In our post about Failing Slow, Failing, Fast, and Failing Very Fast, we discussed sequence risk in the context of retirement planning.
Portfolios with the same returns over the long run can behave very differently when regular deposits or withdrawals are occurring.
In the tournament this year, there was a wide range of ranks in the individual rounds.
Even the winning portfolios all ranked at the bottom of the pack for some rounds. Likewise, the portfolios that finished near the bottom of the pack still had some stellar individual rounds.
Brackets like dsmulonas ranked better than even their best individual round rank while brackets like jannchi and mburgess ended up worse than their individual round ranks.
Thinking ahead to next year, there are some ways of awarding brackets that are consistent as a bonus for having lower sequence risk. That sure would have been good for brackets like Pete_N that saw 5th, 2nd, and 1st place finishes in the last three rounds of the tournament.
Benchmarks can be a very tricky subject to pin down. Choosing different ones can create drastically different backdrops to frame both short and long-term results. This is especially important for setting expectations going forward.
In our March Madness Bracket Challenge, the value-weighted benchmark took the winning prize. However, this portfolio was essentially just a systematic allocation method. If you were not solely focusing on the undervalued (underdog) teams, it probably was not your appropriate benchmark.
After the returns are tallied, you will always be able to find an investment that did better than yours. Seeing the final results can be counterproductive in this way.
As investors, we must constantly battle behavioral biases such as hindsight, anchoring, and confirmation.
Managing these biases isn’t that important when competing in a one-off tournament like a bracket pool, but when we have constant opportunities to alter our portfolios based on current information, avoiding the compounding effect of our own behaviors can be one of the biggest determinants of long-term success.
Thank you to all who participated in our 2018 March Madness Bracket Tournament this year. We hope you’ll participate next year and share it with your friends! The final results are below.