Index | Recent Threads | Who's Online | Search

Posts: 100    Pages: 10    Prev 1 2 3 4 5 6 7 8 9 10
Last Post
New Thread
This topic has been viewed 6084 times and has 99 replies
InspectorSector
Re: Improvement of my Ranking System

ROE should work much better. But when you say "works much better" you need to make sure you are testing against 7+-year-old stocks only. This means creating a universe of stocks that has 7 years of data minimum, then run your ranking system on it. Otherwise the newer companies could be influencing your results. How? First of all there are IPO lockups that cause stocks to tank 6-9 months after IPO. Second, smallcaps have been underperforming for several years. So if you include these newbees in your universe, they are going to be placed at the bottom of your ranking system, assuming the formula comes out as "N/A", making the RS look better than it should.

SteveA

Dec 28, 2019 12:30:14 PM       
Nicoletta
Re: Improvement of my Ranking System

Thanks a lot to all (SteveA, Yuval, Georg, and Jerome). It's a plasure to write here and find experienced people that helps you.

It's sure not my intention to create a RS with a single rule. I am seeking new way to improve my poor results, trying each factor and assembly the one with better results in an unique RS, and I just wanted to know if there's a way to avoid the drop in the last decile that some parameters have.

As far as I understand the answer is no... no way to filter...unless to change the formula.
As Steve says "You should use it in conjunction with other factors. Use enough good factors and the top end roll-off that you see in the ranking performance should disappear"
Thanks again
Fabio

Dec 29, 2019 10:27:08 AM       
yuvaltaylor
Re: Improvement of my Ranking System

Thanks a lot to all (SteveA, Yuval, Georg, and Jerome). It's a plasure to write here and find experienced people that helps you.

It's sure not my intention to create a RS with a single rule. I am seeking new way to improve my poor results, trying each factor and assembly the one with better results in an unique RS, and I just wanted to know if there's a way to avoid the drop in the last decile that some parameters have.

As far as I understand the answer is no... no way to filter...unless to change the formula.
As Steve says "You should use it in conjunction with other factors. Use enough good factors and the top end roll-off that you see in the ranking performance should disappear"
Thanks again
Fabio


Abs (FRank ("factor", #sector [or #industry or #all], #desc) - 80 [or 70 or 60]) with lower values better

This is, in my experience, a good way to use factors for which you don't want the highest or lowest value in a ranking system. Alternatively you could specify an actual value that you think a stock should aspire to. For example, I use the following formula in my ranking systems:

Abs(LoopMedian("OpMgn%(Ctr,TTM)-OpMgn%(Ctr+4,TTM)",12,0)-1.5)

That means I want companies whose operating margins have grown between -5% to 8% per year over the last three years. 1.5% is the midpoint there. Of course, this makes much more sense as a screening rule, but I like putting everything into a ranking system for the sake of flexibility.

Yuval Taylor
Product Manager, Portfolio123
invest(igations)
Any opinions or recommendations in this message are not opinions or recommendations of Portfolio123 Securities LLC.

Dec 29, 2019 2:32:30 PM       
InspectorSector
Re: Improvement of my Ranking System

Abs (FRank ("factor", #sector [or #industry or #all], #desc) - 80 [or 70 or 60]) with lower values better

The only thing I don't like about this is that you can end up throwing the baby out with the bathwater. I prefer to solve the problem by throwing more factors into the mix. If you have enough factors the best stocks should remain near the top.

SteveA

Dec 29, 2019 3:39:19 PM       
wwasilev
Re: Improvement of my Ranking System

Abs (FRank ("factor", #sector [or #industry or #all], #desc) - 80 [or 70 or 60]) with lower values better

A great technique when there's a supporting theory for down ranking the extremes. For example, for dividend paying stocks, avoiding those with high yield since they may have higher risk of dividend cuts and also avoid low yielding issues since they can fail to impose fiscal discipline on management;
abs(Yield-FMedian("Yield",#Industry)) // Lower is Better


But I wouldn't do the same just to patch-up a poorly performing factor. Seems like data-mining to me.

Walter

Dec 29, 2019 4:06:24 PM       
Edit 1 times, last edit by wwasilev at Dec 29, 2019 4:12:52 PM
yuvaltaylor
Re: Improvement of my Ranking System

Abs (FRank ("factor", #sector [or #industry or #all], #desc) - 80 [or 70 or 60]) with lower values better

The only thing I don't like about this is that you can end up throwing the baby out with the bathwater. I prefer to solve the problem by throwing more factors into the mix. If you have enough factors the best stocks should remain near the top.

SteveA


That depends on the factor, Steve. There are a lot of factors for which it simply doesn't make sense for the top values to be best and the bottom values to be worst. You have to think each factor through logically.

Change in margin is one such factor. The companies whose margin has risen the most have decreasing sales and increasing income. That is not a good thing.

Sales growth is another such factor. Companies with unsustainable sales growth are usually punished by the market. I bet you could look at a hundred years of data and see this play out well.

Size may be a third such factor. Really really really tiny companies may not outperform in the long run, especially if you take transaction costs into account. While the SMB factor works pretty well in theory, you have to place a limit on the S (and this may depend on the holding period).

I would guess that about a third--maybe even half--of the most commonly used factors are NOT linear in terms of the slope of the bucket performance but instead follow a skewed bell-shaped curve, and there's usually a good reason for that drop off. It's absolutely vital to think every factor through. Ask yourself: Are stocks that have super high factor X really more likely to outperform than stocks that have middling factor X? You'll often find yourself answering: Well, not necessarily.

And while it may be "data-mining" to estimate an optimal FRank or value for a factor, that may be more LOGICAL than settling on the highest/lowest value.

Yuval Taylor
Product Manager, Portfolio123
invest(igations)
Any opinions or recommendations in this message are not opinions or recommendations of Portfolio123 Securities LLC.

Dec 30, 2019 10:00:32 AM       
Edit 1 times, last edit by yuvaltaylor at Dec 30, 2019 10:02:01 AM
rtelford
Re: Improvement of my Ranking System

Good discussion here folks.

I agree that it's important to see how multiple factors work together, however I’d have to agree with Yuval's point above, many factors are mean-reverting, i.e. stocks with a particularly high factor score may be at the peak of their cycle for the given factor and can be positioned for a downswing, indicated by the drop in performance of top ranked deciles.

Another possible source for the drop-off in top ranked deciles is outliers. An exercise I find very helpful is to screen the stocks for the given factor and take a look at the actual values for the factor across the stocks, not just the overall rank chart.

I’ve actually been looking at this recently in the Magic Formula (and hopefully a Seeking Alpha article to follow). Take Return on Capital from the P123 Greenblatt ranking system:

OpIncAftDepr(0,TTM)/New Working Capital or ( NetPlant(0,qtr)+ Recvbl(0,qtr)+ Inventory(0,qtr))

If you create a screen report with this formula, you will find the median ROC values for the top 30 stocks today is 19%.

The stock ranked as #10 is INVA, a pharmaceutical firm. Its ROC is 365%, a significant outlier from the rest of the portfolio. Maybe there’s good reason for this, however it’s worth digging deeper.

INVA currently has no net plant or inventory assets on its balance sheet, so the numerator in the ROC equation will be quite small, resulting in a very high ROC value, and also a high Magic Formula rank.

Is a ROC value of 365% predictive of future returns? As INVA is an asset light company, ROC is likely not a good factor to use for INVA and may not be predictive of future returns as the Magic Formula intended.

For what it’s worth, INVA lost 24% since being purchased by the Magic Formula model one year ago.

This is one specific example, but the point is that it may be worth taking a look at the actual values of the stocks being screened by your factor(s), outliers like the above may also be skewing your decile ranking performance results.

Cheers,
Ryan

Dec 30, 2019 3:39:12 PM       
InspectorSector
Re: Improvement of my Ranking System

Change in margin is one such factor. The companies whose margin has risen the most have decreasing sales and increasing income. That is not a good thing.

We must be thinking of two different definitions of margin Yuval, or perhaps you are thinking of a specific industry that behaves differently than the norm. Increasing margins shouldn't have anything to do with decreasing sales, in fact increasing sales generally improves efficiencies and hence margins. If you truly want to "think this through logically" then perhaps the answer is to find additional factors that weed out the stocks that are prone to mean reversion. Otherwise, you are indiscriminately throwing out the baby.
Sales growth is another such factor. Companies with unsustainable sales growth are usually punished by the market. I bet you could look at a hundred years of data and see this play out well.

So there are certain companies with unsustainable sales growth AND there are also companies out there with very sustainable sales growth. I don't see that as a reason to discard all high growth companies. In fact, I have systems that capitalize on high sales growth that target specific industries.

To each his own.

Size may be a third such factor. Really really really tiny companies may not outperform in the long run, especially if you take transaction costs into account. While the SMB factor works pretty well in theory, you have to place a limit on the S (and this may depend on the holding period).

I've never been a size-based trader so this one I don't care about.

Dec 30, 2019 4:44:39 PM       
yuvaltaylor
Re: Improvement of my Ranking System

Change in margin is one such factor. The companies whose margin has risen the most have decreasing sales and increasing income. That is not a good thing.

We must be thinking of two different definitions of margin Yuval, or perhaps you are thinking of a specific industry that behaves differently than the norm. Increasing margins shouldn't have anything to do with decreasing sales, in fact increasing sales generally improves efficiencies and hence margins. If you truly want to "think this through logically" then perhaps the answer is to find additional factors that weed out the stocks that are prone to mean reversion. Otherwise, you are indiscriminately throwing out the baby.


Margin is earnings divided by sales. There is no other definition of margin. If a company's sales go down and its earnings stay the same, the margin goes way up.

Yuval Taylor
Product Manager, Portfolio123
invest(igations)
Any opinions or recommendations in this message are not opinions or recommendations of Portfolio123 Securities LLC.

Dec 30, 2019 4:57:31 PM       
Edit 1 times, last edit by yuvaltaylor at Dec 30, 2019 4:58:10 PM
InspectorSector
Re: Improvement of my Ranking System

If sales go up then efficiency improves, better utilization of resources, etc. Thus, when sales rise margins rise. The other option for rising margins is cost-cutting which often occurs when the company's sales have stagnated or they are losing money, or greedy. It shouldn't be too difficult to discern the different scenarios.

Dec 31, 2019 3:08:57 PM       
Posts: 100    Pages: 10    Prev 1 2 3 4 5 6 7 8 9 10
 Last Post