Limits of Fama-French Regressions with Small Industry Samples
Summary
The document considers whether monthly data for a small set of Swedish stocks can support Fama-French portfolio construction and regression. It warns that sorting a limited universe by characteristics may leave only one or two stocks in some portfolios, creating small-sample bias and weakening the significance of results. It also cautions that US factor data may not represent a Swedish industry; global factors may still fail to capture the characteristics of a small local market.
For an industry representation, the response suggests using equal weighting or market-cap weighting rather than optimizing the portfolio. It recommends broadening the asset universe or considering factors and markets better matched to the research setting. The advice is qualitative: it identifies representativeness and sample size as constraints, but provides no empirical comparison or procedure for correcting inference when the available Swedish sample remains small.
Key ideas
- Small stock universes can leave characteristic-sorted portfolios with too few constituents for reliable inference.
- A portfolio intended to represent an industry can use equal weighting or market-cap weighting.
- US Fama-French factors may not describe a Swedish industry or market well.
- Broadening the asset universe or matching factors to the market can improve representativeness.
- The discussion does not offer a statistical correction for results based on a small sample.
Tags
Full text
# Regressing using Fama-French portfolios with small amount of stocks # Regressing using Fama-French portfolios with small amount of stocks I'm doing some research for my thesis and I was wondering if it is possible to only use monthly stock price data for 22 stocks and construct Fama-French portfolios out of them and then regress? What worries me is that the portfolios created will not be "diversified", may this be a problem and what could a potential solution be? I can only access this small amount of stock data as I want to see how well it can describe a particular industry on the Swedish exchange. ## Answer by Alex Bădoi (score 1) https://quant.stackexchange.com/a/25398 If you want a portfolio to accurately describe an industry you should not perform any kind of optimization on it. You have 2 options: - equal weighted portfolio - market cap weighted portfolio Search for S&P Index Methodology for an example and yes, you have a small sample bias issue. i am guessing you want to make factor portfolios sorted on size , one on value etc ? Then you might end up with 1-2 assets in some portfolios. this is a problem you can not solve and it will affect the significance of your findings, whatever you are looking for. ALSO very important is FF data is USA only so if you use it in Sweden it would be a very incorrect approach, unless you use global factors but then you are restricted to 4 factors including momentum - that will not be representative of the tiny market that exists in Sweden. My only suggestion is to look at a bigger market such as USA (you can use FF 5 factor model(2014) or EU market (3 factor + charhart's momentum) as a whole where you will have many assets to pick from. Ideally you need 50-100 assets in each portfolio. I suggest using S&P 400 , S&P500 and S&P600 together. This will cover like 95% of US market cap across a wide range of firm characteristics - and u can use the 5 factor model Alternativelly use the DAX and other EU indexes. also if i may sugest. there are 2 other factors which i use Bettign Against Beta and Quality Minus Junk. you can find data here , the papers you can google yourself. good luck
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.