Slippage study - Smallcap/microcap R2G port

Models
1

Further to the R2G weekly slippage problem, this new thread is a continuation of my comments on the “Slippage – FYI” thread in Simulations and Portfolios section

The previous thread showed an example of success with Bayesian slippage on a large cap value-oriented R2G port. Refer to that thread for my definition and description of Bayesian slippage. This is an example of a failure, a small cap/microcap R2G port.

This was my first experiment with Bayesian slippage. This port is maddening because buys often open higher on Monday than Friday’s close…and head higher all day, and the reverse is true of sells.

BUY slippage study, based on 90 trades from May to Oct 2013:

Average slip % at Monday open: plus 0.70 %
Average Monday low: minus 0.82
Average Monday close: plus 1.51
Average Friday close (Friday to Friday): plus 2.95
Average change from market price (average of highs and lows) to Friday: plus 1.68, 100% invested by end of week
(As you can see, about 1/2 the gains for the first week for this R2G port are on Monday)

“Best” Bayesian slippage for the week: plus 1.22% (on entire port, including cash position) - ~ 78% invested by end of week

Monday limit order: minus 1.3% - resulted in 31 of 90 stocks bought
Tuesday limit order: minus 1.0% - 17 buys
Wednesday limit order: plus 0.4% - 15 buys
Thursday limit order: plus 0.5% - 6 buys
21 stocks left unbought

If one has the cash ready Monday morning, it looks like buying at market (1.68) is better than any combination of limit orders during the week (1.22). The sole advantage of the Bayesian approach is it does leave the R2G investor with some cash over the weekend to employ Monday morning.

SELL slippage study, based on 74 sells from May to Oct 2013:

Average slip % at Monday open: minus 0.32 %
Average Monday high: plus 1.31
Average Monday close: minus 0.51
Average Friday close (Friday to Friday): minus 0.95
Average change from Friday close to market on Monday (average of highs and lows): minus 0.41%
(On average the sells for this port tend to open the week low, head lower on Monday, and head lower all week)

“Best” Bayesian slippage: minus 0.16%

Monday limit order: plus 1.4% - resulted in 26 sells of 74 stocks
Tuesday limit order: plus 1.8% - 12 sells
Wednesday limit order: plus 10 (yes, ten)% - 1 sell
35 of 74 sold at close on Friday for an average discount of minus 2.27% (I might have done better if I hadn’t picked Friday’s close, but it was convenient to work with)

The one outlier, 10%, has a huge effect on the overall slippage. As an example, change it to a more typical 1.8% and Bayesian slippage becomes minus 0.33%

IMHO, for this particular R2G port, Bayesian limit sell orders (0.16 – 0.33) are no better than market sell orders (dump the stock at Monday’s open for an average slip of 0.32). The market order approach also puts one in cash immediately.

Using a more thorough computerized approach may improve upon my manual Bayesian study, which uses limited amounts of data and trial-and-error on a spreadsheet

2

You say ““Best” Bayesian slippage for the week: plus 1.22%”. What do you mean by this in layman’s terms. Limit order of Fridays close plus 1.22%? on Buys
and
On Sells, Friday Close minus 0.16%

Or do I have the plus and minus backwards?

3

Hi rallan:

Your studies are very enlightening. I can only imagine how much work this has taken you.

Thanks for sharing.

Brian

4

What we need are simulation of limit order. If we had it, then every trader could research and simulate buy-limit and sell-limit at various distances from previous close and for different time duration (GTD limit order) to find the right order to enter or exit for each of the systems we trade. Of course some assumptions need to take place and the simulation would work best for stock that trade some volume, hence the probability of getting partial fills would be negligable, but in any case it would add tremendous value to P123 usability. As an important side benefit - limit orders bypass the thorny problem of slippage. This feature was requested many times before, maybe its time for it to get the priority it deserves.

It has come up repeatedly in forum discussions, here are few:
http://www.portfolio123.com/feature_request.jsp?view=open&cat=-1&featureReqID=37
http://www.portfolio123.com/mvnforum/viewthread_thread,2854#12815
http://www.portfolio123.com/mvnforum/viewthread_thread,7016#34823
http://www.portfolio123.com/mvnforum/viewthread_thread,6062#28864

Thanks,
Zvi

5

Hi Rallan,
while the above studies are enlightening (thanks for all this work), I think it is mostly another example of curve fitting.
When we all know what the “ideal” limit orders were in the past, they will most likely be different going forward.
For R2G ports crowding and overlap recommendations will eat up any advantage, this approach might have.

For bigger, non public ports, I can see some justification for this approach though.

6

“Best” means this is the best I could do to improve the performance of the port, with trial-and-error, applying different limit orders on different days of the week.

For buying:

Suppose I have $100,000 in cash to invest in this R2G port Monday morning. It’s 6:30 am, I receive my email from port123 indicating which stocks to buy and which to sell. I look up what prices the stocks closed at on Friday. With my online account I place my limit orders:

Friday’s closing price minus (or less, if you prefer) 1.30% of those prices i.e; $10 Friday close means $9.87 Monday buy limit order (therefore my bid price is at a discount) on ALL buy recommendations, end up buying a few of the recommendations,

…then Monday night place an order equal to Friday’s (not Monday’s) closing price minus 1.00%, i.e; $9.90 Tuesday limit order, for Tuesday and buy a few more,

…then Tuesday night place an order equal to Friday’s (not Tuesday’s) closing price plus 0.40% (I must now start offering a premium) i.e; $10.04 Wednesday limit order, for Wednesday and buy a few more,

…and so on.

At Friday’s close that week, the $100,000 port will be, on average, worth $101,220 (an overall 1.22% increase) of which ~22% is cash.

If all stocks were bought with a market order (mid-point Monday high and Monday low) on Monday instead, hypothetically the port would be worth $101,680 (an overall 1.68% increase) by Friday’s close at the end of the week, and 100% invested.

For selling:

So, on Monday morning I enter into my online account Friday’s closing price plus 1.4% of those prices i.e; $10 Friday close means $10.14 Monday sell limit order (therefore my asking price is at a premium), on ALL sell recommendations, end up selling a few of the recommendations,

…then Monday night place an order equal to Friday’s (not Monday’s) closing price plus 1.8%, i.e; $10.18 Tuesday limit order, for Tuesday and sell a few more,

…then Tuesday night place an order equal to Friday’s (not Tuesday’s) closing price plus 10%, i.e; $11.00 Wednesday limit order, for Wednesday and sold one more,

…and so on, eventually selling whatever is leftover at market on Friday, at the end of the week, for an average discount of 2.27%, on the week, for these leftover stocks

With this Bayesian approach if the port was worth $100,000 Monday morning, it’s value drops to $99,840 (0.16% slippage from sales) or to $99,670 (0.33% if I adjust for the 10% outlier in my data), by the end of week.

If sold at market right at opening Monday, it drops to $99,680 (0.32% average slip) by close on Monday. For the hypothetical Monday market sell order (midpoint high and low) this is $99,590 (0.41% average slip)

7

Too much work!

8

Your point is well-taken. Necessity is the mother of invention; I had to address the problem of too much slippage using the tools I had available to me and the circumstances I found myself in, this is what I came up with. I am trying to get beyond blind guessing.

More work needs to be done to optimize limit orders for R2G ports to minimize slippage.

But for this port, I couldn’t fit a curve, at least a satisfactory one.

To re-iterate, it’s easy to conceive of a port where this approach will never work i.e, for buying, the price opens high and heads steadily higher all Monday and all week. Obviously, best approach is to buy at market on Monday. This may describe most of the small-cap/microcap R2G ports.

I don’t doubt the weightings of optimal limit orders for stocks drift, just as the weightings of the Port123 stock valuation factors drift in a market cycle, or go to pieces like they did during the 2008 financial crisis. Right now, I am the only one who knows exactly which ports I am talking about. With R2G pay ports, the # of subscribers is limited so any recommended prescription for setting daily limit orders would also be restricted to the subscribers.

The “overlap” business may not be as bad as you think in regards to prescribing an optimal set of daily limit orders for a port. The different R2G ports tend to make different recommendations on different weeks, I have had opposing buy and sell recommendations by different ports for the same stock the same week, a different vexatious problem with R2G :frowning:

Going forward, if a set of optimal, that is best, daily limit orders for each R2G port was validated by backtesting, this validation should be re-validated and up-dated in real-time going forward, a form of out-of-sample testing.

As an aside, I suspect the longer the average holding period for the stocks in a R2G port is, the less important a Bayesian approach, and for that matter limit orders, is. Ben Graham and Warren Buffet never talked about limit orders.

see the thread “Slippage - FYI”