A Gratuitous Deconstruction Of The StockRanks

9th Apr '15 by Edward Croft

56 comments 25566 reads

When we launched the Stockopedia StockRanks back in early 2013 we had optimistic but cautious expectations. It was a ranking system formed using fundamental and technical factors that have always paid off in stock markets. But as every good quant knows, even the best designed strategies can underperform for significant stretches of time. Whether we'd get off to a good start or not was, as Ragnar Lothbrock might say, up to the gods.

So while the direction of the StockRanks to date hasn't been entirely unexpected, the extent of the outperformance has been rather fortuitious. The 55% return from the top 200 ranked stocks over the last two years significantly exceeded our own internal expectations.

Being of Scottish origin, I've always suffered from downside paranoia. So over the Easter break, aside from spending some lovely time with family, I indulged in a gratuitous analytical deconstruction of the StockRanks. I hoped to gain a better understanding of their nature and the confidence to continue investing in them.

Rules of the hunt

“You certainly usually find something, if you look, but it is not always quite the something you were after." J.R.R. Tolkien, The Hobbit

In designing a ranking system for stocks, you know what you are trying to find… in our case we are searching for good, cheap, improving stocks - but what comes out the other end of the factory may not be quite what was being searched for. After all, if stock markets are efficient at setting prices, then good quality firms should be more expensive than poor quality firms, so trying to find good and cheap companies ought (in theory) to prove impossible !

But I've always thought theory is for the birds, efficient markets are bunk and stock markets are there to be beaten. I want absolute conviction that after we've applied our ranking process we are actually finding the archetypal kinds of good, cheap, improving stocks we set out to, not some wishy-washy mash of stocks where each positive factor has cancelled each other out.

So what I've done is the following:

Sort the entire UK stock market universe by ascending StockRank.
Chop up these 2000 UK stocks into 10 equal sized buckets (i.e. from StockRank 0-10, 10-20, 20-30… all the way up to 90-100).
For each StockRank bucket calculate a set of average ratios…

Unlock with Email

Unlock with Google

Already have an account?
Login here

cig

10th Apr '15

A couple of challenging periods for the Stocko strategy would be:

- Momentum crashes (e.g. 2007): this tend to be rare but violent, though the QV component will tame the blow.

- Speculative bubbles (e.g. dotcom): QV stuff tends to be left in the dust compared to indices in those periods, though this is more of a relative thing: you need to be content with your hairdresser doubling their money in between haircuts, while you're staying flat.

Both of these are probably part of the explanation why these relatively well known anomalies have so far persisted long term, along with the behavioural aspects Ed mentions. Many investors will just bail out during those crisis, taking a loss at the worst time.

On tradability, the Stocko strategy can be declined to all caps and is pretty stable (e.g. does not require too frequent rebalancing), and works long only, so no major issue here. May not be as good as the paper account but should do okay with a sensible rebalancing strategy. One thing Stocko could do is to have some actual company petty cash (retail-sized) invested in a mechanical variant and publish the net performance. Much easier to do than an ETF, and could provide useful data points (and perhaps some insights as to best practical rebalancing rules), as well as have some marketing potential along the "money where our mouth is" theme.

Mark Simpson

Hi Ram,

Nice idea but I'm not surprised this effect is randomly distributed - sample size is very small, timeframe is very small and you are looking at post change stockrank. You really need to have a bigger sample size and look at the stockrank at the beginning of the quarter not today because what you are really interested in is how likely a high/low stockrank stock will see a big rise in the future not what is the current stock rank of previous big risers. Also the implication that a high number of +30% risers will generate good returns excludes the chance that the 50-60 decile might just be more volatile and have had 50 shares that were -30% fallers giving a poor average return. So a good analysis would have mean, volatility, skewness etc. not just no. of stocks >x%.

If you took a snapshot of the stockrank of all stocks now and did the full analysis in 3months then that might tell us something, but TBH this is better done by someone who has access to point of time data in the database. Ed?

Cheers,

Mark

WarrantStar

Cig,
Thanks for your reply. I guess it may be too early to tell what the effects might be, because we don't really know a lot of the details yet. The people running the fund would want to make it perform as well as possible, so as to out perform other funds and draw in as much money into the fund as possible. So it would be very tempting to base the trades on Stockrank information which the rest of us would not have had access to yet. It the fund included lower market cap shares then this could affect their share prices.
If the situation was that the Stockranks were published on here at the same to the fund managers and the rest of us, then theoretically that might give us some advantage. But I don't think it would be a fantastic opportunity. If it was a fantastic opportunity then a few hundred people on here would be seeking to exploit it, thus making the opportunity smaller and harder to access.
So my current view is that the launch of an ETF could have some negative impacts, which may be mitigated by having very clearly enforceable rules designed to prevent the fund managers having earlier access to Stockranks information than the rest of us.

Edward Croft

Ram - rather than looking for 30%+ winners, the more interesting thing is to look at the hit rates at different cutoffs in each basket 0%, 5%, 10%, 15% etc. I've got stats on this but they have always shown since launch that there are far higher hit rates in the top baskets. The probability of picking a winner has been higher (so far) in the higher ranked buckets.

I've got analytics in the backtester and we'll definitely publish them at some point - but they are v slow to compute so they can't be live at the moment.

roddy10

I also note that stockranks is QVM - but there is also a ranking by QVGM - it would be interesting to compare the two?

@pka - We haven't changed the StockRanks since launch. We've fixed a few bugs along the way, but the factors have remained static. We are though researching a major adjustment for 2016 - though I'm loathe to change what is a working model.

What interests me more is allowing subscribers to 'fork' our ranks. There are plenty of approaches which would and could do far, far better than the StockRanks have to date. Actually the top 1% or momentum rank stocks that are investable, rebalanced quarterly, has returned over 85% since launch - there's a myriad of approaches.

@randomambler We are working on a portfolio automator which I hope will help do a lot of what you are looking on.

@roddy10 Quant strategies are definitely tradeable, but of course nothing works all the time. One of the benefits though of incorporating momentum into a value strategy is that the length and depth of underperformance can be minimised versus normal value strategies. QV strategies like Greenblatt's can suffer for much longer periods of time. It's worth reading Novy Marx on this stuff.

@cig - yes momentum crashes are tricky, but they get a particularly bad hit due to the collapses on the shorting side. e.g. in 2009 the top shorted momentum shares were the high balance sheet risk value stocks (Grahamesque) and they absolutely rocketed back in the rally - so the shorts got pummeled and the long/short momentum trade 'crashed'. The long momentum stocks though didn't suddenly die at the time - on the contrary they were full of safe low-beta stocks, they trucked along fine but underperformed the rally as they hadn't fallen as much. I do agree wholeheartedly though that a long-only momentum strategy can be a hard to implement strategy due to churn and volatility.

@roddy10 QVGM has underperformed QVM to date on all rebalancing schedules... see below... (blue line is the QVGM, green line QVM).

I think there are a few reasons for this.

Firstly growth is highly followed by the market, and thus over-prized. As a result it tends to mean revert. It doesn't perform as well across the deciles as other ranks (e.g. the 60-70 rank segment has done the best). This is backed up in our own tests as well as in the academic literature.

Also the Growth Rank is correlated with the Q & M ranks, so it actually doesn't add any new information - as highlighted in the article above the StockRanks are very exposed to growth already.

Thirdly, if you have a 4 way rank, the stocks become less exposed to Value, which is a strong factor. QVM is 33.3% Value, whereas QVGM is only 25% value. That impacts the returns too.

Ed - many thanks - that is really really interesting.

This is looking at outliers to judge a model that predicts averages. Most of these large short term move outliers are likely to have been news-driven, something the Stocko model doesn't, and cannot, take into account. So it should be pretty uncorrelated with Stockranks, which this sample is consistent with.

PhilHofGrowYourDough

Hi Ram,

In your exercise you're considering the Stock Rank now (as in today) ... for me a more useful measure is the Stock Rank 3 months ago, i.e. do stocks with a high stock rank consistently outperform those with low stock ranks.

After all what we are discussing here is "If I pick stocks with high StockRanks how confident can I be about their performance relative to lower StockRank selections?"

rather than "Can I identify a stock that will rise 30% in 3m and drive itself towards a high stock rank?"

My own personal position is that I select high stock rank shares and watch them closely as the price rises (pruning out those that don't) and whilst holding I'm quite comfortable for the StockRank to drop (usually as the ValueRank decreases) as long as the price momentum continues.

I currently hold 20 stocks.
Using the European&US version of Stockopedia (as I hold UK, US and EU stocks) ...

8 stocks rank 98 - 99,
6 rank 90 - 97,
3 rank 80 - 89,
1 ranks 74,
1 ranks 69
1 ranks 57.

Over the last 3m only two stocks have achieved > 30 %. One has a StockRank of 98, the other 57. Over the last 3m my portfolio has increased by ~8%. Not stellar given the general rise in the markets but ok, however the portfolio performance over the longer term has been very strong.

So I guess a few points from my ramblings:

1) For me a high StockRank is like a shooting star. You might only see it once for a particular stock in your investment lifetime. If the stock meets my other criteria and I have cash to buy it I take the plunge and I stick with it (regardless of fading StockRank) until the medium term price momentum disappears. That is why I hold two stocks with mediocre StockRanks namely Sprue Aegis (LON:SPRP) - StockRank 69 (Gain +389%) and GFT Technologies (ETR:GFT) - StockRank 57 (Gain +142%).

2) For me StockRank now (as in today) isn't related to past performance (although the Momentum aspect contributes close to a third of the StockRank) rather I use it to reduce risk (don't buy duffers) and maximise performance going forward (with the knowledge that some of those selections will not work out)

3) Three months and 30% might not be the most ideal parameters to use when considering the effectiveness of high StockRank.

Regards
Phil

Ramridge

Hi Ed -
I will go along with that, although I am generally wary of too much slicing and dicing. Not seeing the wood for the trees, etc.

OK. Couple of people have picked up that I am using today's StockRanks rather than StockRanks at the start. I accept that and it bothered me. So I have gone back and picked out the value of StockRanks as at 8 Jan 2015. Basically this can be done by clicking on the Print drop down and choosing the pdf for 8 Jan. A tedious exercise that I am not in a hurry to do again.
The chart now looks like this

Possibly slightly worse than before.

I said in my original post that this is not statistically significant as the sample size is small, etc. But for my purposes, the conclusions remain unchanged.

Hi Ram,

Really tedious exercise. I appreciate the effort.

I wonder what would happen if you did the same for stocks that have fallen > 30% over 3m?

I think this might be the aspect that pushes the 90th percentile to the fore.

Let's face it we all know there are big gains to made in very speculative plays but they also come with big risks.

Warm regards
Phil

Yes - Ram I think you are measuring volatility rather than anything else over just 1 quarter.

To answer PhilH's question... the number stocks that fell > 30% in Q1 2015.

0-20: 31
20-40: 36
40-60: 17
60-80: 9
80-100: 3

I did that with a £10m cutoff.

Bearbull

19th Apr '15

@Ed
Don't know if I still get an answer: According to the introduction the deconstruciton study bases on 2000 UK Stocks. The FTSE All Share consists of 635 Shares including 136 Investment Trusts. AIM50 consist of 50 UK companies. So what about the other 1450 UK shares? Are these all very small companies?
James Montier criticized that most of the time smart beta ETF's just capture the well know small cap effect.
How about your deconstruction study? Does it not just show a big percetage of small cap effect and a little percentage of QVM Input? What if you would run the same study only on FTSE 350 shares ex Investment Truts. Would the result remain as clear cut as it is now?

Actually there's a negative correlation between StockRank and size. If you analyse the average StockRank of each decile, the lowest tiers of the StockRanks are skewed towards microcaps, while the higher StockRanks have a higher weighting of mid caps. Interestingly, the top 10% are on average smaller than the 50-90 range.

Here's the chart:

5533fa702ccc1Cursor_and_StockRank_Decile

Depends on what you call "very small" companies. See http://www.stockchallenge.co.uk/ftse.php to get a feel of the distribution of company size and FTSE membership. A fair number of AIM companies are larger than the smaller main market small caps.

28th Apr '15

@cig
Thank you for the link. Unfortunately this link does not really answer my question as only 1209 UK shares are listed on this stockchallenge-web-page and not the 2000 the study bases on.
@Ed
On the Stockopedia Directory site the title-list says that there should be 2637 UK Shares. 2526 shares listed at the LSE and 111 shares listed at ICAP Securities Exchange.
First of all, when I download the pages 1 to 27 into Excel I can only count 2056 shares.
2637 – 2056 that means 581 shares announced in the title bar are not in the database? mistake?
Second, in the deconstruction study you state, that you use the “entire UK Stock market univers” and further that the study used 2000 UK shares.
The value of any study respectively of a screening depends a lot from the definition of the univers of stocks that you include in a study. If you use 2000 stocks but let’s say 300 of them are investment trust, 50 of them are a different share-type or listing-type of the same company and 500 are not investable as the share price is about 0.5 to 50 pence or there is no liquidity then the result of the study is blurred or spongy and at the worst misleading.

Often a detail concerning the universe of shares that is used for a screening study makes a big difference in the usefulness of the result. For example Joel Greenblatt mentions in the footnote on page 166 in the appendix of his "little book that beats the market" that he excluded utilities and financial stocks for his screens. Now, in the UK about every seventh company is a financial stock. That means the magic formula will not lead to an optimal result in the UK market.
Back to the deconstruction study: I still would appreciate to have some more information about the structure of the 2000 shares that were used for this study. How many shares are constituent of which index(es) (I cannot get this information form the Directory Page) How many of them are liquid enough so that you can invest £1’000 without pushing the price 10 to 20% away from the last traded price? So any further information or link would be welcome.

I ran the numbers in spreadsheets using exports from the screener - they are perfectly replicable if you have the time to do the work.... though we do plan on integrating these studies into the website as a whole to save everyone the bother.

How many stocks? I overestimated in my article... after filtering down from the 2000+ in the directory, the actual list was 1,617 in total. This is the complete number of our own own classification of primary listed 'screenable shares'. The checks which reduce the number in the directory include:

Exclude dual-listed stocks, no GDRs etc.
Exclude investment trusts,
Exclude non-trading securities
Exclude those with no trading in the last 90 days.
Exclude companies that are inactive / not currently filing statements
Exclude companies with no fresh statements for the last 18 months
Exclude those with no price histories (you'd be surprised how many there are)

How many shares are constituent of which index(es) ? We include all the shares on LSE and ISDX. I'm not going to break it down further as that includes all the FTSE indices... i.e. FTSE 100 - probably 100 etc.

The lower StockRank stocks are heavily weighted towards micro caps as can be seen above. It's no surprise - they tend to be low quality stocks with poor price momentum - by definition it is to be expected.

How many of them are liquid enough so that you can invest £1’000 without pushing the price 10 to 20% away from the last traded price? No idea right now... that is something that's dependent on market dynamics, market maker liquidity and all kinds of variables - I don't know is the answer. I'm sure we could do a spread analysis too, we'd likely find that the lower StockRank stocks have far higher spreads, so trading costs would be higher... not sure who'd want to own them though.

Hope that helps.

6th May '15

@ Ed I thank you, Mr. Croft, for your replay. I hope my questioning was not too fussy. In my view, your information about the pre-filters was valuable.

A gratuitous deconstruction of the StockRanks

Rules of the hunt

Unlock the rest of this article with a 14 day trial

About the Author

Edward Croft

56 comments

Unlock the metrics for all these stocks and more...

Unlock the metrics for all these stocks and more...