Backtesting

[NormR]

I expect to spend the day doing some backtesting on a Bloomberg tomorrow. 

 

I'd be happy to run a quick backtest for those who don't have access to such a device.  Just propose a stock screen that I can check out. 

 

The overall structure tends to be ...

1. pick a universe (i.e. common stocks on the NYSE with market caps in the top 70%)

2. specify some combination of stock picking criteria (i.e. lowest 10% of universe by P/E and then lowest 10% of the resulting list by P/B)

3. select a rebalacing period (i.e. monthly, quarterly, annually with annually being most sensible for value investors)

4. select a portfolio weighting scheme (ie. equal or market cap weights)

5. pick time frame (i.e. the last ~20 years in the U.S. and ~17 years in Canada)

 

 

[Gopinath]

Thanks for the offer Norm, Here is a set of criteria that I am curious about:

 

1. Universe (all over the world)

2. ROE > 12%

3. Debt/Equity < 35%

4. Insider Ownership > 10%

5. P/E <10

6. Time frame - As long as possible, Re balancing (1-3 years)

[vikx01]

Thanks Norm. Here's a list to try.

[*]Universe = US and Canada, Market cap between (50M and 1B)

[*]Criteria = lowest EV/EBIT quartile and then highest and positive 12 month price momentum quartile

[*]Rebalancing = quarterly

[*]Weighing = Equal weight

[*]Timeframe = As long as data is available.

 

 

 

[NormR]

Thanks for the offer Norm, Here is a set of criteria that I am curious about:

 

1. Universe (all over the world)

2. ROE > 12%

3. Debt/Equity < 35%

4. Insider Ownership > 10%

5. P/E <10

6. Time frame - As long as possible, Re balancing (1-3 years)

 

I went back 20 yrs / annual rebalancing.  I'd take the result with a grain of salt due to small stock issues, etc. 

 

CAGR:  20.8%

[NormR]

Thanks Norm. Here's a list to try.

[*]Universe = US and Canada, Market cap between (50M and 1B)

[*]Criteria = lowest EV/EBIT quartile and then highest and positive 12 month price momentum quartile

[*]Rebalancing = quarterly

[*]Weighing = Equal weight

[*]Timeframe = As long as data is available.

 

Note that the mcap criteria isn't adjusted for inflation in the back test.

 

17 Years ending 12/31/16, CAGR: 20.7% in CAD

[vikx01]

Awesome! Thanks Norm!

 

 

Did you do any backtests of your own that were interesting and that you could share here?

[Gopinath]

Thanks Norm!

[NormR]

Awesome! Thanks Norm!

 

Did you do any backtests of your own that were interesting and that you could share here?

 

I probably did 100+ yesterday.  But I've yet to do a postmortem on them, calculate CAGRs, etc.

 

A large part of the work is trying to figure out where things "break", doing sensitivity tests, etc.  As a result, much of it isn't particularly interesting on its own when taken out of context. 

 

A few general observations.

 

Classic value ratios applied to the large U.S. stock space generally haven't worked very well over the last 20 years.

 

Simple techniques often work better than more complicated ones.

 

It's easy to run into weird oddities in the data that can skew the results.  This happens with any database to some extent.  But it takes awhile to figure out where the potholes are.

 

One fun one I did yesterday was a quick check on an overly bold assertion that I've made in the past.  That is, doubting the efficacy of buying junior mining companies based on low-P/B alone.  It's next to madness I figure ;)

 

Anyway, I looked at metals & mining stocks on the TSX for the 17 years through 12/31/16.  A market cap weighted portfolio of the sector generated a CAGR of 4.2% and reaffirmed my prejudice against stocks in the sector.  (My apologies to all the good miners out there.)

 

The bottom decile (10%) of the sector on P/B (equally weighted, annually rebalanced) sported a CAGR of 21.2%.  The result had me laughing out loud. 

 

The portfolio came with a standard deviation of 46%.  It generated lotto like returns every few years and it also suffered from huge crashes.  Practically speaking, the gains were driven by tiny stocks.

 

Not a practical strategy, IMHO, but I'll have to stop slagging off low-P/B junior mining stocks  ;)

[Hielko]

With those kind of small stocks it's questionable if those returns can be archived in real life. Bid/ask spreads can be huge, and liquidity very low. Without actually trying to buy the stock you never know at what price you can in and out.

[vikx01]

 

Norm,

Thanks for sharing the results from your backtest study of M&M on TSX.  M&M's have had higher SD compared to the TSX. I am assuming the results free of survivorship bias.

 

I'd love to hear about your other screens.

 

 

[NormR]

With those kind of small stocks it's questionable if those returns can be archived in real life. Bid/ask spreads can be huge, and liquidity very low. Without actually trying to buy the stock you never know at what price you can in and out.

 

Very true, the old roach motel type stock.

 

[NormR]

 

Norm,

Thanks for sharing the results from your backtest study of M&M on TSX.  M&M's have had higher SD compared to the TSX. I am assuming the results free of survivorship bias.

 

That's the claim.  But, in my experience, the data can be quite shaky when it comes to nano-cap stocks.  I usually do my best to screen out such situations when backtesting.