I have three unrelated questions

[Whycliffes]

I have a system divided into 7 factor themes, with around 150 nodes, where some nodes are weighted up to 10% and some down to 0.2%

I have noticed that when I test the ranking system in the simulator, even though I use 25 stocks in the portfolio and a test period of 20 years, with the US and Canada as the universe (5500 companies), very small changes in the weighting of the nodes can result in large percentage changes in returns. A change in weight on a node by 0.5% can cause the entire simulator to produce 3-5% worse returns (from 49% to 45%), although I tested it on a portfolio of 25 stocks.

Is that normal? I would have thought that since the portfolio is so large, and divided by the number of nodes and factor themes, as well as the period in which it is tested, it would be a little less sensitive?

I’ve been looking a bit for ways to reduce slippage. I see there are some that recommend VWAP orders when I trade in stocks with low volume, but if I don’t use market orders and only place limits at yesterday’s close (or close) until the order is filled, will there be anything to be gained on TWAP orders? (maybe I have misunderstood what the VWAP order dose)

The reason I ask is because that type of order is a lot more expensive to trade, and it can only be used for one day at a time anyway.

Until now, I’ve split the portfolio 50/50, between the U.S., Canada, and Europe, but how do you guys distribute?

There is no doubt that historically, the US has been the best over the past 15 years, and in the simulator tests I do, the US provides much better returns. So even though I achieve pretty much the same alpha in Europe and Canada, the total yield is much better in the U.S because of the overall market.

But we don’t know anything about the future; it may be overfitting - and a bad idea - to distribute the portfolio based on past history?

[yuvaltaylor]

1. If you’re just looking at CAGR%, yes, just choosing or not choosing one or two super-performing stocks during one or two small periods will make a huge difference in overall returns. That’s why it’s much better to trim your returns first and then measure your alpha. If you do that, you shouldn’t find huge differences when you change weights a little. The easiest thing to do is to take all your one-week or two-week or four-week returns, delete the top and bottom 2% or 5%, and measure the alpha of the rest. (Alpha is very easy in Excel: it’s simply the intercept of your portfolio returns on the benchmark returns.) I use elliptical trimming, which is a lot more complicated, but the result isn’t very different.

2. A VWAP order submits small orders throughout the day. Here’s an example. Let’s say I want to buy $30,000 worth of a stock that normally trades only $15,000 per day. If I submit the whole order, the price is going to move a lot, and not in the direction I want. But if I submit fifty small orders of $600 each, that may not move the price much at all. If you place a limit order at yesterday’s close, there’s a good chance it won’t be filled. You should use limits with VWAP orders too, but it’s better to use a wider latitude than yesterday’s close. Here’s something that happened to me the other day. I wanted to buy a bunch of shares in PKOH. The open was about 4% higher than yesterday’s close. I placed a VWAP buy limit order under the open and resigned myself to the probability that it wouldn’t get filled, particularly because the spread was close to 2% of the price. But at about noon the price drifted under my limit and then the small orders began to be filled; by the end of the day I had filled my entire order at a price well under the open.

3. Personally, I put between 20% and 25% of my portfolio into European stocks and the rest into US and Canada. But I didn’t decide to do that based on past history. Because the US has outperformed Europe so thoroughly over the last fifteen years, a strong case can be made that the reverse will happen over the next fifteen. The main reasons I keep my portfolio mostly in the US and Canada are a) I have access to Compustat data for US stocks and not for European stocks; b) European stocks are harder and more expensive to trade. It has nothing to do with performance.

[rwbattyaz]

I’m going to take a crack at question #1 with some basic math. With a universe of 5500 companies being ranked the interval value for 1 company is 0.018181%. Assuming you are buying the top 25 companies that is the top 0.454546%. Assuming a 2X RankPos sell rule this means you are working with the top 0.909090% of the universe.

Given a 0.5% weighting change this is really 1.0% since there is both an add and a subtract somewhere else in the system. And this 1% is larger than the 0.9% you draw from the universe.

With the universe size and relatively small portfolio holdings from it you are very sensitive to “small” changes. Options are to cut universe size, use screening rules to restrict result set, increase portfolio size, extend sell rules, or some combination.

Cheers,
Rich

[Whycliffes]

@yuvaltaylor

A VWAP order submits small orders throughout the day. Here’s an example: Let’s say I want to buy $30,000 worth of a stock that normally trades for only $15,000 per day. If I submit the whole order, the price is going to move a lot, and not in the direction I want. But if I submit fifty small orders of $600 each, that may not move the price much at all. If you place a limit order at yesterday’s close, there’s a good chance it won’t be filled. You should use limits with VWAP orders too, but it’s better to use a wider latitude than yesterday’s close.

My lower limit is to only trade 1/10th of the average monthly dollar volume.

Will the very placement of a limit order of 30 000 move the market because the market sees that in a stock with an average of 15 000, there is a large buy order?

What I usually do is place a buy-to-limit order at yesterday’s close, and after an hour or two, if it is not filled, I place it above today’s low, and then two hours before closing, I place the order at what is the buy limit (buyerside) at the time. Do you think that a VWAP order still will give better fills?

[rwbattyaz]

@duckruck Not my main point which is given the problem parameters it is inherently unstable with respect to small changes. The boundary that has been drawn between “in” and “out” is quite small. Using a two dimensional analogy this is the red dot in the middle of a dartboard.

In regards to portfolio size my approach is to run an ensemble of small strategies with each taking a different slant on identifying holdings. I am currently running 7 strategies, sized from 5 to 11 holdings, totaling 55 positions with 5 duplicate positions. I just benched a 4 position strategy because it had become totally Energy. This approach is high turnover with the lowest being about 10X which implies a more liquid universe. So 25 positions is not a lower limit.

Cheers,
Rich

[yuvaltaylor]

@Whycliffes

What I usually do is place a buy-to-limit order at yesterday’s close, and after an hour or two, if it is not filled, I place it above today’s low, and then two hours before closing, I place the order at what is the buy limit (buyerside) at the time. Do you think that a VWAP order still will give better fills?

Yes, I do.

[abwillingham]

I recently started using vwap with “no slippage” via the P123/IB interface and I love it.
I can set it on multiple accounts and go for a walk.
It may not get filled that day but the next day I click a button and it tries again.
Its far less time consuming than what I was doing.

[Jrinne]

TWAP or time-weighted average price orders have not been discussed yet and may be worth considering for some people or for some stocks.

I generally use TWAP (time-weighted average percent) 5%. I am selecting stocks that I expect to have increasing prices and positive returns over a relatively short period. Waiting all day (or waiting several days) should, on average, result in purchasing a stock at a higher price if my is model doing anything useful at all with regard to predicting price changes. And even if not, the market is generally increasing.

The sims suggest that buying at the close has the lowest returns compared to at the open or average of hi, low 2X close. And delaying a day or 2 really reduces the returns of my sims dramatically.

TWAP 5% does not have that much market impact according to my reading. I DO NOT HAVE ANY PERSONAL DATA TO SUPPORT THAT. So I just discussing my reasoning without any personal data to support my method…

FWIW, I basically always use TWAP 5% (when not placing a market order). I have no definite proof that it is a good idea. But it may be worth considering for some–at least for the stocks that are moving away from you aggressively with high volume based on news or a very significant analyst upgrade, for example.

Jim

[abwillingham]

@Jrinne

TWAP or time-weighted average price orders have not been discussed yet and may be worth considering for some people or for some stocks.

Jim, I would love to try TWAP but that does not seem to be a supported algo in P123.

[Whycliffes]

@yuvaltaylor

Because the US has outperformed Europe so thoroughly over the last fifteen years, a strong case can be made that the reverse will happen over the next fifteen. The main reasons I keep my portfolio mostly in the US and Canada are a) I have access to Compustat data for US stocks and not for European stocks; b) European stocks are harder and more expensive to trade. It has nothing to do with performance.

Yes, here is the performance over the last 20 (10) ( years in some of the markets:

[AlgoMan]

These are index performance in the local currencies, can be a bit misleading.

[Whycliffes]

@yuvaltaylor

You should use limits with VWAP orders too, but it’s better to use a wider latitude than yesterday’s close.

How do you use it? What price do you normally set as a limit in your VWAP order?

[yuvaltaylor]

About 2% below yesterday’s close for a sell and about 2% above for a buy. But it depends on market direction: if it’s a generally bullish day I’ll go higher and if it’s generally bearish I go lower.