TestFol.io

A few days ago a reader suggested I check Testfol.io and…I am so glad he did! (another guy on Twitter later told me that their database is not always correct, so ¯\_(ツ)_/¯).

When it works, Testfolio is the best of PortfolioVisualizer with the best of PortfolioCharts. PV was constrained by the financial history of each specific product in its db; but when you take $SPY, it is pretty silly to consider that you can analyse it only since the launch of the ETF. $SPY is just the S&P500 Total Return index (if you reinvest the dividends) minus management expenses. Extending the “life” of $SPY to the whole life of the index, again correcting for expenses, shouldn’t be that hard.

That’s what PC does (and more. But here that’s not relevant). The issue with PC is that it covers only 3 asset classes: stocks, bonds and gold (I find commodity beta, in its plain vanilla form, only useful to understand that you should not invest in commodity plain vanilla beta). PC factor selection is also limited, letting the user only play with value and size. PC is a blunt instrument in a world (especially for investors that can access US-listed securities) with an ETF for basically any strategy.

The most annoying thing about PC for yours truly is the lack of any trend following proxy/strategy.

This is where Testfolio had me at hello: I can finally test trend on a multi-decade horizon. Trend strategies and related funds have been around for 40 years, the longest track record might belong to Dunn Capital Management – a CTA founded in 1974. Testfolio has two extended trend instruments:

• KMLMX: in 1988, Mount Lucas Management introduced the MLM Index as the first passive index of returns to futures investing.  The objective of the MLM Index strategy is to provide pure systematic trend following exposure in a consistent, efficient, and cost-effective manner which captures the price risk premium offered by those who seek price certainty. $KMLM is the ETF that tracks this index and KMLMX is the index minus the ETF management fee (0.9%): this allows Testfolio to extend the backtest period from 2020 (when the ETF was launched) back to 1992 in a reasonable manner.
• DBMFX: $DBMF is an ETF designed to replicate the SG CTA Index at a lower management cost. The SG CTA Index is built by pooling the performance of the 20 largest and most liquid “trend following funds”, creating the closest thing to a benchmark for the strategy. These funds charge HF-like fees, the infamous 20-and-20, therefore DBMF tries to compensate for its model’s lack of sophistication with a low-cost implementation:

• Considering that $DBMF is so far delivering on its promise, DBMFX is a rough approximation of how $DBMF should have behaved before its launch. The instrument is built from the SG CTA Index performance, plus 2.5% (the fee of the underlying funds), minus 0.85% ($DBMF management fee). There’s obviously no guarantee that the future, nor the past, would/would have perform(ed) accordingly.

How I built my backtest

I wanted to keep things simple for this first iteration. Here some salient points:

• VTSIM is Vanguard $VT ETF with an extended backtest horizon and no management fees. I choose $VT instead of $SPY because…well, if you do not understand why, go back enjoy your Bud while measuring stuff like we live in the Middle Ages. The fact that there is no management fee is a bit annoying and I realised it after I did the analysis but…$VT today charges 7bps, so not really worth re-doing everything.
• IEFTR is $IEF with more history and 15bps management fee. Self-explanatory.
• KMLMX over DBMFX: it allows a longer backtest, that’s basically it. IRL I suggest a blend-model approach, to use more trend ETFs in the same trend sleeve to diversify model risk. If you think that KMLM has a ton of survivorship bias, good, we are two. But…trend has proven to work in many ways. HF managers themselves concur that anyone, as in…any retail normie investor, can build a simple model trading 4 markets and get 80% there. The future would not look like the KMLM-included past but it would not massively differ either (the biggest difference would probably point to what vol-target your trend strategy/ies aim to).
• CASHX: here we see the second massive improvement offered by Testfolio compared to PV. We can easily adjust the management fee of each instrument to our desired level. Increasing CASHX management fee by 30bps allows the backtest to mimic the real-life cost of leverage more closely. 30bps over 3m T-Bills represent the cost of leverage embedded in futures (including a safety margin), the cost any retail investor using Return Stacked ETFs (as an example) would pay. A margin loan from IBKR is obviously more expensive than that.
• the three ingredients are equally weighted. Not the greatest analytical approach, uh? Based on what I recall from the thousands of hours listening to podcasts, the ideal configuration should be something like 45 trend, 40 stocks and 15 bonds. But if I think about risk parity, the opposite solution comes to mind. So EW and call it a day; not that we are dealing with the most rigorous dataset anyway, so…

Here is the result:

Adjusted for inflation:

Almost 6% REAL with 15% max drawdown and sub-8% vol. This is just to give you a very rough idea of the type of Safe Withdrawal Rate you can achieve by mixing these 3 ingredients.

If you want to match a 100% stock return:

The drawdown chart should put to rest any 100%-stock portfolio stan. As a reminder, the IRL strategy uses 2 ETFs: $RSSB and $KMLM equally weighted, rebalanced once a year. I am pretty confident my wife can manage that, with a 30-minute explanation. Not that I think it is the optimal way to implement it, just evidence that today’s ‘complexity’ ain’t no more the reason why you should not take this road.

I already highlighted before the survivorship bias potential issue. Let’s not forget that the benchmark here is the best-performing index of the best-performing asset class of the last century. If you want to apply a margin of safety on how the future might look like compared to the past, you should do it for both strategies.

Unsurprisingly, the $SPY went through some cycles, i.e. the lost decade we all know about:

What is more worrying, for me, is that this model-strategy 60-month CAGR is in a clear (?) downtrend. It started above 15% and then declined to around 5%.

In this backtest, I used $KMLM instead of $DBMF. The above picture confirmed what I already knew: trend also had its lost decade, basically right after $SPY’s one ended. If 15% CAGR was probably due to all three ingredients occasionally being in sync, 5% might be due to a back-to-back sequential weakness of trend into bonds.

these CAGR are inflation-adjusted

30 years is a long period but it is long enough to draw a meaningful conclusion? Decade-long regimes can easily warp results in one direction or another. Higher nominal (and real) risk-free rates should improve bond performance going forward while trend “alpha” (leverage costs – collateral income) mean reverts, as it has been tentatively doing for a couple of years.

Maybe the ‘real’ long-term Sharpe of the stock/bond/trend combination would settle below 0.7 but I would definitely take the over compared to a 100%-stock portfolio. [I refer to the Sharpe ratio here because, as I hope I have demonstrated, anyone can easily dial up/down leverage to get whatever target CAGR they want…and risks would move consequently]

What I am reading now:

Follow me on Twitter @nprotasoni