Is value investing dead?
Warning - lots of words and numbers to follow.
I've mentioned before that I use my software development skills to help me with investing. I thought I'd share what I do and how I go about it.
A topic that's always been of great intrest to me was backtesting based on fundamental analysis. There is already a lot of great literature on the subject, such as "What works on Wall Street" by James P. O'Shaughnessy and "What has worked in investing" by Tweedy, Browne. Ben Graham and Seth Klarman were also proponents of semi-automated stock picking strategy, among many others. Reading all these works have been of tremendous benefit to me, and is the backbone of my investment strategy. I have always felt, however, that there are some unanswered questions. They may lead to the same answer, but I always wanted to ask. Such as:
1. Most studies base strategy on buying at a particular point in time, then rebalancing at a particular point in time. Usually going from lowest to highest. That is, backtest on Low EPS strategy would rebalance portfolio based on stocks with lowest PE once a year, or so. This would be fine for a fund manager who can accumulate all stocks meeting criteria. Not so fine for a smaller investor such as myself, that will probably buy only a few stocks and not once a year, but when an opportunity comes up.
2. Not many studies based on Australian stocks.
3. Knowing how a strategy works across an entire market is very useful, but even more useful is knowing how it works across stocks that I personally am interested in. For instance, I don't invest in mining and biotec specs, startups with no earnings history, companies with chronic debt/acquisition problems, etc. This is very personal and no automoated filter can be setup for this. Basically I want to backtest just on stocks I am interested in, this is very specific to my style/situation.
4. What about portfolio management? Some things that may make a big difference are averaging down/up, selling losers, limiting number of purchases per month/year.
5. What if you have limited capital? What percentage to allocate to each purchase? How long to hold for? When to sell? What if I want to combine multiple criteria, and possibly have different ones for buy and sell?
Overall, I wanted a system that would allow me to setup an algorithm that would step through, historically, day by day, and make buy/sell/topup decisions as I would, in real time. At the time, I couldn't find anything that met the criteria, so I built my own.
The first test of the system, was, of course, the classical Ben Graham approach - buying companies at below asset value. Does that approach still, statistically, make money, even though it's been known about for many decades? Below are my ramblings on it.
DISCLAIMER: This post is for discussion only. This is not a system I use, nor do I recommend it to anyone. There's lots of things that could throw an error here. It is based on my own list of ~230 companies, filtered based on nothing but my own opinion, there's some survivorship bias, there may be bugs/data errors. I repeat, this test is done on a very specific set of data. Doing it with a different data set is likely to produce different results.
For the purpose of this exercise, all backtests will be:
a) Run for the last 10 years, 01/11/2003 to 01/11/2013
b) based on ~230 companies in my watchlist. These include some companies that have since been delisted, but not all, so there is some survivorship bias. Most of these are companies with at least a few years of generally profitable operations.
c) no dividends included
d) $10,000, or 2% of portfolio parcels, whichever is greater. $30 brokerage, based on closing price, no slippage.
e) backtest will step through on a monthly basis. Each week, it will evaluate current holdings to see if any meet the sell criteria. It will evaluate all companies to see if they meet the buy criteria.
First step is to establish a benchmark.
XAO returned 69.4%.
The proper benchmark, however, is simply buying all the companies that I will use in the backtest and holding on to them. This returned 158.04%, and will be used as the benchmark.
Criteria for Buy will be Market Value compared to Total Assets. I will break this down into multiple backtests to show the results based on discount size.
Criteria for Sell will be Current Market Value > Net Assets X 3.
No averaging up/down.
A strong correlation between asset value and return, especially at the lower end. Buying at below 0.25 P/B shows massive returns, but there were only 26 trades over 10 years. Given that I am already using a cut down data set, this simply isn't enough data to treat as reliable evidence. Nevertheless, my results showed what other studies have shown over the last few decades - buying at lower book values tends to give better returns. And the lower you buy, the better. What's also interesting is that the success rate has not changed by much in each test, if anything, it improved with lower value P/Bs, which would normally be considered riskier.
Let's now take it further. For buy criteria, I will use P/B < 0.70, as that is the value at which the test above returns above the benchmark, at 236.41%. Time to play with sell criteria:
There seems to be relatively little difference whether to sell at PB>1 or PB>5. However, selling sooner would have tax implications, which are not accounted for here. Also, my capital is limited and I do not have enough funds to buy everything that meets the criteria. So let's make changes to accomodate me. Capital will be restricted to $250,000, 2% per trade, so that is only enough initial capital for 50 trades. Because we now won't be able to buy everything that matches that criteria, companies wll be ordered by P/B. The lowest ones will be chosen. This should give us an idea whether it is better to recycle the funds, or stay invested once bought. Let's now re-run the tests above with these changes.
Tests above confirm that it's better to sell holdings before they become too expensive, but just at what price is a lot less clear. The higher you go, the more capital is tied up for longer. Also, at P/B > 5, there's only 49 trades, as most of the capital was tied up in the first few years and no cash was available to buy opportunities available later.
My take on it is that this is one of the many things that can't be automated. Selling at P/B of anywhere between 1 and 4 would make sense, depending on the number of new opportunities available and cash needed.
Let's go on.
I'll use P/B < 0.7 as buy criteria and P/B > 2 as the sell criteria for future tests. P/B > 3 shows a slightly better result, but I prefer a lower one as it results in more trades, and the result is less likely to be attributed to a lucky few trades.
On of the more surprising things to me was that the success rate was consistently above 70% for all tests. Nevertheless, I would think it prudent to pick safer companies, with better metrics. Even if it doesn't improve the result, it should lower risk. So let's try a few things to see if we can improve the result, such as net debt/equity, ROC, OCF. Admittedly, this is not a good test for this data set, as it is already mostly made up of more solid companies.
Amazingly, none of these show any positive contribution to either risk or return. But again, it must be stressed that my data set is already heavily skewed towards "safer" companies. Checking for positive operating cash flow over 5 years gives a slightly better result, so I'll use that as the benchmark going forward.
As the success rate is quite high, it would be reasonable to assume that averaging up/down should give a better results. Let's try a few different tests. Original buy criteria will be checked when topping up, no averaging on deteriorating fundamentals in other words.
As expected, averaging does produce a better result. What's less expected is that averaging up produced better result than averaging down. Let's now keep averaging up, 20%, twice, as the new benchmark.
I am limited to how much I can post in one entry, so I will continue in the next post.