Ivo Welch, April 2, 2025

ASAM Attribution and Final Report Guidelines

Our Benchmarking Factors

Note: In the $R M - R F$ , as in the CAPM, RF is not the risk-free yield but the risk-free rate of return! So, the return on VGSH (or VFIRX) is a good substitute. The prevailing short-term Treasury yield is not. (For one, the latter is annualized, whereas you need a same-period return.)
We use daily returns with dividends for benchmarking, because daily stock returns help with 2nd moment estimations.
This gives us a practically easily achievable benchmark factor model.

We do not have three years of monthly data on which to evaluate our strategy. We only have about 4 months.
- We are an educational class, not a fund.
- We cannot help this.
- Our means (averages) suck as estimates of true performance. There is little we can do.
- Our variances suck, but less so.
We should have better estimates of variances and covariances by working with daily returns rather than monthly returns.

If we had invested in factor (=portfolio) X instead, how much would we have made?
How loaded/related is our portfolio to factor X?
What part of our portfolio is “due to” our loading on factor X?

Both your and the factor portfolio can have alpha with respect to other models, such as the CAPM.
It is no shame to have lower alphas when more factors are included. You are really working out how similar your portfolio is.
- You may not want to tell clients that they can get your alpha elsewhere, though,
- but you may also be on the buyer’s side.
Think about it: it is no shame for Vanguard that its VTV value portfolio does not outperform the factor model.

Try different factor models (on the RHS), at least

If $r_{p} - r_{f} = a + b_{p 1} \cdot (f_{1}) + c_{p 2} \cdot (f_{2}) + ϵ_{p}$ then $μ (r_{p} - r_{f}) = a + b_{p 1} \cdot μ (f_{1}) + c_{p 2} \cdot μ (f_{2}) + μ (ϵ_{p})$ where $μ (x)$ means the average of x.
The regression fitting sets $μ (ϵ_{p})$ to zero.
The first factor must be $r_{m} - r_{f}$ , or in our case VOO – VGSH.

The attribute contribution is each term in the sum (beta times mean), which will add up to your portfolio return.
This works with more factors, too.
All factors (here, $f_{1}$ and $f_{2}$ ) should be zero investment pfios, just like your own $r_{p} - r_{f}$ . Recall: $r_{f}$ is VGSH.
If we use a long-only portfolio as the basis of a factor $f$ (as we do in ASAM), then we must subtract $r_{f}$ to make it a zero-investment portfolio.
We should use daily returns to have better exposure estimates.

Is your portfolio performance due to loading on a particular industry? Use industry funds.

Where to get industry funds? Search!

Examples:

In the real world, different funds use different benchmarks and attributions.
For example, if you are a Tech software fund, you may want to use the VGT Vanguard fund as your benchmark.
- For example, VGT (Info Tech) - RF could be one…or energy, or….
Pick your own factors and tell us where your performance comes from.

Because you were targeting the SR, you should know how to calculate it, too.
Also, calculate it on a factor-replicant of your portfolio.
Remember that the SR is $μ (r_{p} - r_{f}) / σ (r_{p} - r_{f})$ . The $r_{f}$ should be small.

The typical outline should be:

Investment Objective
- Sharpe ratio, not Alpha
Explored Trading Strategies
- Intuition (why do you think market neglected your choices?)
- Academic Literature
Implemented Trading Strategies
- Reasonable halfway English, halfway precise description
Ex-Ante Marketing Pitch
- based on backtests
Performance Analysis
- see above in these notes
Learning Experience
- what would we do different today?
- where did we screw up?
References
Appendix
- Include sample calculation of metrics.
- Describe portfolio in more detail (stocks, etc.)
  - give example calculations, providing the key metric for one or two of your input selections
  - describe sorting if multidimensional
- Must be sufficient to it possible for other team to replicate portfolio selection