Why Long-Short Portfolios Help Test Equity Factors
Summary
The document explains why factor research often compares stocks with high and low values of a characteristic. A typical method sorts stocks at a point in time, then compares the subsequent returns of the high-characteristic group with those of the low-characteristic group. The return spread corresponds to a long position in one portfolio and a short position in the other. If returns vary monotonically with the characteristic, this spread can make the relationship easier to detect statistically than examining only one side.
The answer distinguishes testing a factor from implementing it as an investment. Investors should inspect how much each leg contributes to total performance; a strategy may be less attractive if much of its return depends on the short portfolio. A short leg can also be understood as a negative active weight relative to a benchmark. The example describes momentum portfolios: the high-momentum group outperformed the market while the low-momentum group underperformed, so the combined long-short result reflected both effects. Historical portfolio returns illustrate the point, but do not establish that the factor will persist or be practical after trading costs.
Key ideas
- Factor tests often sort stocks by a characteristic and compare subsequent returns across high and low groups.
- A long-short spread can improve the statistical visibility of a monotonic relationship.
- Long-only results do not capture the low-characteristic portfolio’s contribution to a factor spread.
- Investors should evaluate the return contribution of each leg before implementing a factor.
- The momentum example shows both high-group outperformance and low-group underperformance contributing to the spread.
Tags
Full text
# Fama-French Long-Short portfolios. Is short really necessary?
# Fama-French Long-Short portfolios. Is short really necessary?
I am student here who just touched on Fama-French portfolios. I read from certain articles that while fama-french portfolios are extremely popular in explaining the returns, in real-life, it is quite impractical to short due to transaction costs etc. Hence, no managers out there are offering fama french portfolio as an investable product.
My statistical knowledge is not very strong im afraid. But i am just wondering why fama french portfolio studies require long-short (or they call it zero dollar portfolio)?
Or some books i have read say that to test a potential factor for example an ESG Factor we can also take the top decile stocks returns with low carbon emission minus the bottom decile stocks returns with high carbon emissions to create the factor portfolio.
What would be the implication if only go with long-only portfolio without doing any shorting? e.g., I would long all the small cap companies, and no shorts on large cap companies. would it have any repercussion to my research on factors?
please let me know if my question doesn't make sense. But im just wondering why all the factor portfolios studies out there are having a short leg requirement
## Answer by Enrico Schumann (score 2, accepted)
https://quant.stackexchange.com/a/78679
The standard methodology for a single "variable" (such as price-to-book) is this: 1. Sort the stocks in the cross-section at a specific point in time by that variable. 2. Compare the subsequent returns of stocks that have a high value of the variable, with returns of stocks that a low value of the variable. To do this comparison, you'd take the difference between the returns of high-value-of-variable stocks, and returns of low-value-of-variable stocks. This, in turn, corresponds naturally to being long one portfolio of stocks (high-value-of-variable), and being short another portfolio (low-value-of-variable).
Statistically, if there is a monotone relation between the variable and subsequent stock returns, the spread in returns will be greater if you look at long/short portfolios, and so the power of your tests increases. (That is, the relation between variable and return is easier to spot.)
That being said, as an investor, you will want to analyze to what extent long and short positions have contributed to total return. For instance, if you find that the short portfolio was responsible for a substantial portion of total return, then not implementing it would mean the strategy becomes less attractive. (But of course, it need not be an actual short position. As pointed out in the comment by nbbo2, it might be a negative active weight relative to a benchmark.)
Kenneth French publishes several datasets that show the returns of the original sort portfolios, i.e. not just the long--short aggregate returns. For instance, momentum (i.e., the variable you sort on is 1-year return). The figure below shows the total returns of 10 momentum portfolios since 1970, with 10 being the high-momentum portfolio and 1 the low-momentum portfolio (the lighter the grey, the higher the momentum). In blue, I have added the market return.
```
library("NMOF")
library("zoo")
library("PMwR")
P <- French(dest.dir = tempdir(),
dataset = "10_Portfolios_Prior_12_2_CSV.zip",
weighting = "value",
price.series = TRUE, na.rm = TRUE)
P <- window(zoo(P, as.Date(row.names(P))),
start = as.Date("1969-12-31"))
M <- French(tempdir(), dataset = "market",
price.series = TRUE, na.rm = TRUE)
M <- window(zoo(M, as.Date(row.names(M))),
start = as.Date("1969-12-31"))
greys <- grey(seq(.1, .6, length.out = ncol(P)))
par(las = 1, mar = c(3,5,1,2))
plot(P <- scale1(P, level = 100),
plot.type = "single", log = "y",
xlab = "", ylab = "Performance",
col = greys)
lines(scale1(M, level = 100), col = "blue", lwd = 2)
par(xpd=TRUE)
text(x = max(index(M)), y = c(coredata(tail(P, 1))),
labels = 1:10, pos = 4, col = greys)
par(xpd=FALSE)
```
As you can see, the high-momentum portfolio beats the market, while the low-momentum portfolio underperforms. The headline "momentum-return" number aggregates both sources of outperformance.Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.