This is the third post in a mini-series on how factor returns from academia can be different to those from smart beta investing. For the previous post on how factors tend to work better in small caps, click here.
Sir John Hawkins was an English ship’s captain during the 1500s, and was the second cousin of Sir Francis Drake. Despite being poorly educated, Hawkins made his fortune by being the first Englishman to profit from the ‘Triangle Trade’. He would buy textiles and rum in Europe, and sail them to Africa, where he would sell them in exchange for African slaves, which he would then transport to America, selling them in exchange for sugar, tobacco, and cotton, which he’d then transport to Europe, selling them for textiles and rum – continuing the cycle and making a profit on every trade.
Hawkins made a fortune from the Triangle Trade, rising to the rank of Admiral in the Royal Navy, and earning himself a knighthood along the way. Despite his great financial success, he was essentially a middleman – facilitating trading between one party who wanted to sell, and another who wanted to buy.
Today’s middlemen
In spite of the modern-day push for disintermediation, especially within the traditional banking sector, middlemen are still everywhere in our society. AirBnB and Uber were founded on the premise of connecting buyers and sellers, and being a middleman can still be incredibly lucrative – just ask Jeff Bezos.
These middlemen also exist in the investment world, connecting those who want to buy stocks to those who want to sell. These middlemen require payment for their services (in the form of a bid-ask spread) which is a cost to both parties wishing to transact.
When it comes to factor investing, these trading costs are another one of the reasons why academic factor performance is unlikely to be replicated in the real world – because trading stocks costs money. The typical Fama-French factor portfolios assume no trading costs, which clearly isn’t going to be representative of a real-world investing strategy.
Exactly how much extra it costs to trade is dependent on the factor. Some factors, like size and value, have low turnover (their strategies don’t require much buying and selling), and some, like momentum, have high turnover (their strategies require quite a lot of buying and selling).
Researchers from Research Affiliates have tried to quantify exactly how much it may cost to implement traditional factor portfolios in the real world, and found that costs do indeed vary based on the turnover of the strategy:
Source: Research Affiliates
Whilst the authors caution against taking these absolute figures as gospel (momentum might not cost exactly 6.1% a year to trade), they do note that relatively speaking they sound sensible – on average, momentum is likely to cost 12 times as much to implement as a small-cap strategy.
Because of this, they find that mutual funds which sell themselves as ‘momentum’ funds (those which include the word ‘momentum’ in their name), have heavily underperformed over the last 25 years:
Source: Research Affiliates
These momentum funds were the worst-performing category of funds in terms of outperformance relative to the market. On average, they underperformed the market by 2.2% to 4.3% (depending on the fund weighting method used). These funds yielded an average −2.6% CAPM alpha and −3.1% four-factor alpha (using the Fama–French three-factor model plus the standard momentum factor).
In other words, the investors in these funds experienced a 3.1% annualized average shortfall relative to the performance of the paper portfolios their factor loadings were replicating. If these funds had been able to fully capture their factor premia, they would have outperformed the market by roughly 0.9% a year (3.1% better than their 2.2% average shortfall).
This finding is consistent with research conducted by academics Novy-Marx and Velikov, who show that almost no factor, constructed as a long–short portfolio, with turnover exceeding 50% has any return left after accounting for transactions costs.
So real-world transaction costs clearly have an impact on factor performance.
Alpha Architect continue the research on trading costs with an excellent article in which the authors do much of the heavy lifting in summarising some other key papers on transaction costs. Given that they’ve done almost all the work already, I’ll briefly go through their findings.
Beginning with 3 papers written by the academic community on the effect of transaction costs on factor portfolios, the research finds that “All three papers above come to similar conclusions — trading costs reduce factor premiums, and momentum, the so-called “premier anomaly,” suffers the most from transaction costs, leaving it with a fairly low capacity and questionable after-frictional-cost performance.”
The papers find that not only do real-life trading costs reduce performance (especially for momentum), as seen in Research Affiliates’ work, but higher trading costs reduce the capacity of the strategy (the amount of money a strategy can invest before performance of the strategy suffers).
Putting numbers to the theory, a key table from one of the academic papers suggests that the momentum factor only has a capacity of $6bn:
Source: Alpha Architect
Turning from papers written by academics to papers written by practitioners (who obviously have a vested interest here), researchers from investment management behemoth Blackrock, using a different dataset, find that capacity for factor funds is much larger than found in academia:
Source: Alpha Architect
Momentum’s capacity has gone from approximately $6bn using one set of data, to over $300bn using a different set.
In addition, researchers from AQR in their paper ‘Trading Costs of Asset Pricing Anomalies’ use yet another dataset, and find the long-short momentum capacity to be around $56bn.
Clearly, small differences in transaction costs models, and the underlying data fed into these models, can make a large difference for capacity estimates.
So which is right?
To identify which dataset is closest to reality, AQR conducted an experiment. They used the trading cost assumptions from the academic research to create an S&P 500 index portfolio and see how much it would cost. What they found was that creating an S&P 500 tracker using those assumptions lead to a portfolio costing 0.63% annually. Given that it costs Vanguard, iShares, and almost everyone else creating S&P tracking products less than 0.10% in costs, it seems likely that the academic estimates of transaction costs are too high.
Conclusion
Attempting to estimate factor trading costs can be difficult and depends on the data and assumptions employed. However, it’s clear that a couple of things can be learned from the research:
- Trading costs reduce performance – any factor strategy which is to work in the real world must pay real attention to the costs of implementation. Higher turnover strategies incur higher trading costs.
- Factor investing strategies do have capacity constraints – higher turnover factors have lower capacity constraints than lower turnover factors, so investors need to judge whether a fund’s size reduces their chance of outperformance
The next post in the series looks at how human factors can influence real-world implementations of factor strategies, and what this means for smart beta investors.