An impressive new R2G model by Kurtis Hemmerling

A frequent contributor to Seeking Alpha , Kurtis Hemmerling made some tweaks to a Piotroski strategy and has made it avaiable on R2G. It’s called Piotroski - Major Exchange Smallcaps

It has very impressive results and, although it targets smaller caps, the Holdings average daily $ amount is around $21M. You can see this new statistic in “Key Stats”

Thank you Kurtis.

Your welcome and I have many other models I want to make available for free to the Portfolio123 community.

Kurits:

Thanks for sharing. Does this portfolio generate its trades completely automatically? Or do you use discretion to exclude some stocks?

Brian

All of my models generate trades 100% automatically. I have 4 more on the way (which will also be free for P123 memebers).

Hi Kurtis:

What liquidity filter do you use for your Piotroski port?

Of the 10 currently held stocks 5 appear to be thinly traded.

CKEP - On Feb 15, it had a total of 4 trades (3 of 200 shares or less and one trade about 3,000 shares)
CVTI - On Feb 15, it had a total of 4 trades ranging from about 100 shares to 900 shares.
I consider the above to be virtually impossible to establish a significant position without moving the price.

CRRC, DEXO and UFCS had about 14-26 trades on Feb 15. These would also be difficult to enter and exit.

The other stocks appear to have decent to great liquidity.

Regards,
Brian

Some are thin unfortunately. My smallcap models are not for everyone. That system requires no over the counter stocks, USA only and minimum $150K per day turnover. But the $150K rule is only meaningful at the time the stock is picked up. Getting in after its been held for weeks or months can have stocks trading at far less liquidity especially when no even is nearby.

My other smallcaps system that has a CAGR of 60%+ uses the following liquidity rule: AvgDailyTot(60)> 200000

Kurtis:

Thanks for the explaination. I appreciate knowing the liquidity settings.

I’ve traded very low volume stocks before. Getting in is easy and stress free – either there is willing seller at a price I like, or I skip the stock and buy another.

But getting out is another matter entirely!

Brian

With all due respect, this is another example of “paper” portfolios/systems.
With many of these stocks trading just a few thousand shares a day, it is impossible to get any meaningful amount of shares. Even if you get them, then the problem is getting out at any reasonable spread.
The problem is not unique to Piotrosky based systems, although this is one of the worst offenders, but also to all those that are based on small caps. When you place a modest requirement for trading volume, say about $ 1MM, they collapse rapidly.
I admire this attempt, but it remains on the theoretical side of the fence.
martin

It is true that these smallcap models are not made for big time traders. But I am not sure that Portfolio123 wants to exclusively target the big boys either.

The model buys and sells at the mid-point of the days range plus 2.5% slippage. This model is a minimum of $150 (entry). Assume you have $250,000 to invest you use Foliofn. Smallcaps makes up 20% of your overall strategy. You follow 3 smallcap systems, each with 10 stocks. You do not need to spread yourself over so many stocks and strategies with higher liquidity models. Thus, you are trading $1,666 per stock. We are only talking about 100 to 300 shares per stock. If a stock is trading at its minimum entry point ($150K), this is little more than 1% of daily turnover.

But if you are trying to put large amounts of capital focused on one smallcap strategy - yah, you’ll get hurt pretty bad with slippage. I tried to put a warning in the model description that this is for small amounts of capital. Personally, I like systems that go places big investors can’t. Lots of alpha there as long as you spread your trades wisely across many strategies and stocks.

I think the point isn’t how much money can be traded in this strategy as much as it is what slippage a small investor is likely to experience and what does that do to the expectations on overall performance. I looked up the stocks listed above in my Interactive Brokerage account and charted bid/ask spread over the past couple days. Below is my estimate of the typical spread over this timeframe:

BKEP: 1.2% (not CKEP as mentioned earlier)
CVTI: 5.0%
CRRC: 0.6%
DEXO: 1.5%
UFCS: 0.6%

Overall the slippage for these stocks would be higher than the assumptions used in the model, but then these were highlighted as being low liquidity. Assuming the other stocks are higher liquidity then perhaps the slippage assumptions are reasonable. Exits are another matter but the model has high assumptions for slippage and low turnover (by my standards). Based on this limited data it seems to me the assumptions behind this model may be reasonable and that it may be tradeable for a limited number of small portfolio’s.

Don

Hi Kurtis,

I cannot see the model. Have you removed it?

The model is the Piotroski model - and its not been too terribly impressive of late. Lots of volatility and a 10% drawdown - it is still up 1.5% from the market since launch but its going through a slump. Most model returns are lumpy so its best to pick a handful that you feel confindent in the underlying theory of alpha-generation and then stick with it. I am most happy with my Value Rockets out of sample performance but who is to say it won’t be next to go through a slow period. I also have a 50 position Value Revision Rockets that I feel quite confident in as more positions means less fitting potential (in my opinion), and I have strong convictions about the blend of techniques being used. 50 positions is a lot…I know…but I want to start widening the range of offerings to different investors/managers.

Kurtis,

Any general comment on the Starmine ARM you site in your documentation for value revision rockets would be much appreciated. Did you work for Stamine?

It looks to me like some of what Starmine does could be duplicated at P123 by identifying whether the most recent upward earnings revision is 1) a movement of an analyst’s estimate from consensus to above consensus or 2) an analyst moving from below consensus to in line with consensus.

Which of the two situations is the case could be determined by determining whether a recent upward revision in the consensus earnings estimate is accompanied by an increase or a decrease in the standard deviation of the estimates going into the consensus estimate. A move from consensus to above consensus by a single analyst will increase the standard deviation while a move from below consensus to in-line with consensus will result in a lower standard deviation.

It might not work if there are multiple analysts making earnings revisions. Or would it? I’d love to see what the buckets look like in a rank performance.

We used their data feeds along with all the major providers. Starmine was impressive and if you are going to build an institutional product properly - its the way to go for sure. You can request their white papers if you like. Their ValMo model was very good and I am a big fan of analyst revisions and surprises so their ARM model was up my alley. Although you get the data feed and white papers of the process - its still sort of black box. Even if it wasn’t, the stuff they do isn’t possible by us. Like they know which analysts have better track records than others and can weight the changes appropiately… they have access to data we don’t and even if we did - they have teams of quants compiling it. But its not really practical to get this sort of data for funds under 50 or 100 million in my opinion. Its just too expensive. It is very easy to be paying 50k per month in just unique or customized data sources.

But testing it and using it as a building block for model creation was enlightening. CIQ is okay (their pre-packed models I mean) but Starmine was much better. CIQ is great for raw data but I personally think the builders at P123 can do better when it comes to model design.

All of them (Zacks, Starmine ICQ) look at “the most accurate analyst”. And all of them put some weight on the most recent estimate: the rest is a big secret. If the most accurate (often the most recent) estimate is higher than consensus they calculate a “Expected Surprise Potential (Zacks)”, “SmartEsimates (Starmine)” etc. Having the most recent estimate (based on the most recent information) above consensus is thought to be favorable for a stock and likely to lead to an earnings surprise.

P123 is better. I suspect P123 can get some idea of what the most recent analyst is doing and be better still. The above method would be completely accurate if there is just one revision in the last X number of weeks.

There may be other better ways. Ideas?

https://dl.dropboxusercontent.com/u/281024033/Starmine/StarMine_ARM_White_Paper.pdf
https://dl.dropboxusercontent.com/u/281024033/Starmine/StarMine_SmartEstimates_White_Paper.pdf

I put a couple of the white papers on my dropbox if you want to read them instead of requesting them from Starmine. Its not just the trading ideas of the various shops and data streams but the data depth in many cases (or in every case where there was huge alpha). For instance, in the one short interest data - there are a couple dozen independent data streams that are compiled and you can isolate them. For example, you can include such variables as the borrowing cost to short (which has great gross alpha but obviously much lower net alpha due to borrowing costs), the difficulty to locate shares, the # of brokers that even have shares, where exactly the borrowed shares are coming from - because if they are coming from insiders and institutions it is a bit more meaningful than if not, and so on and so on. Its very specialized data and not simply a blend of the stuff on CIQ. We had all the CIQ raw data and all of these were in addition to that. And then there is very indepth options data and comparing call and put IV one month out near the money and on again it goes.

But one thing that I found which I was impressed with on a total other note was this site. An academic has been market-timing with out of sample results since 1990. What surprises me is that his site is free, you can see his current recommendation and model - and the record is very impressive. http://www.mojena.com/ I think this is a great double-check for any investor who has his portfolio generating a broad sell signal.

Nice find! What was really impressive was the low frequency of signals.

Kurtis, Thanks. Just saved this to my favorites.

kurtis - do you know if there is any independent verification of mojena’s claimed historical performance? I’m sure he must be honest but I’ve seen too many claims out there that don’t hold up once I start watching the live performance.

Steve

But they do not reinvest dividends at the tracking site so total returns should be higher for buy and hold and mojena system. On the mojena about page you can read about this.