Trading Systems Development

We are proud to bring you a series of articles on Trading Systems Development.  Today we will continue our discussion about ensuring reliability of results.  The last article, if you missed it, was about “The Idea.”

Measurement of Success

Measuring the success of a trading system is more complex than one may initially presume.  Many issues interfere with a straightforward interpretation of results.

Execution Costs

This is a very fundamental cost that must be realistically incorporated into simulations, but it is also arguably the most difficult issue to deal with.  There are several components. 

Fees

Not all brokerages pass through all of the fees. Check with your broker.

• Commissions
  • Include the commissions of the broker that you intend to trade with.  If your commissions are more than 1c per share, you should consider looking for a new brokerage.
• SEC fees
  • Currently, SEC fees are .0000418 times the principle value of securities sold.
• NASDAQ trading fees
  • $0.000075 per share transacted on the NASDAQ.
• DOT fees
  • Trading through OES, it is $0.0005 per share transacted on the NYSE.  Other DOT lines charge more.  Trading through BRUT may be free.
• ECN rebates or fees
  • If you are going to trade NASDAQ, you will likely be trading over ECNs.  You will typically be charged $0.003 for hitting the bid or ask, and will be rebated $0.002 for providing a limit order that gets filled.

Slippage

If a stock last traded at $25.67, and you place a market order, you are probably going to have to pay more than $25.67.  The difference between $25.67 and your execution price is the slippage.  If you place a limit order, you might pay $25.66 instead of $25.67, but you are most likely to be executed on your trade when the stock is going to next trade at $25.65.  The difference between $25.66 and $25.65 would be your slippage.  You are least likely to be filled on your limit trade when the stock immediately moves away from your limit price.

Transactions have a market impact that is generally very transient.  The exact extent of your market impact is dependent on the liquidity of the stock, the spread, and your method of entry.  Larger trades have more market impact, as do trades that take liquidity (hit the bid).  Market impact makes trading more expensive by pushing the price of a stock up when you want to buy and down when you want to sell.

If you were to buy and sell a security randomly using market orders, you would lose money very quickly, even without other transaction costs.

Limit orders also have another cost:  Adverse selection; you are more likely to have your limit order filled when the price moves against you, and least likely to be filled when the price goes away from your limit price.

Adverse selection plus market impact are together slippage.

If you require that your trades be made using market orders, then you should consider the slippage cost to be roughly equal to half of the spread for small orders.

If you are going to provide liquidity (offer using limit orders), reasonable slippage would be between zero and $0.01 for small orders, depending upon how sophisticated your order placement method is.

Providing liquidity is generally cheaper than taking liquidity.  However, you may need to take liquidity if you expect the stock to move very quickly (your opportunity cost is high).  Trading with market on open (MOO) orders for the NYSE is generally very cheap.

Incorrect Benchmarking

It is easy to use an inappropriate benchmark if you have an insufficient dataset, or do not correctly analyze the risks of your stocks.

Survivorship Bias

If your dataset does not include stocks that no longer exist, and you are performing a long-term study, then your stocks will have outperformed the market in general.  Similarly, if you use stocks from a benchmark that has changed their components over time, those stocks will have outperformed.  This holds true for the NASDAQ 100, S&P 500, and all stocks that currently exist.

Fortunately there are several ways to correct for survivorship bias:

• Generate both long and short trading signals, and keep your portfolio dollar-neutral
  • The abnormally high performance on the long positions will be countered by the high performance of the short positions.
• Compare your results to the average performance of the stocks in your dataset, rather than to a market index.

Incorrect Risk Measurement

Expected returns are related to risks.  If you take more risk, you are likely to average higher returns.  There are risk premia for many factors.  A good model is the three factor model developed by Fama and French.  If you do not have book to market or size information, a simple market model may suffice.

Optimization Overestimation

Optimization is a useful tool.  However, there are many problems associated with optimization (most of which we will discuss in future articles). 

When you optimize your trading rules over a dataset, the trading results of those rules in the same dataset will be unrealistically good.  The difference between these results and realistic out-of-sample trading results will be dependent upon the size of your dataset, the number and flexibility of your optimization variables, the number of independent transactions, and the properties of the return distributions in the dataset.

A simple solution to optimization overestimation is to test the parameters your system generates on a dataset that was not included in the optimization. 

Recent Research

An article published in a recent issue of the Journal of Finance present statistically significant evidence showing that stock markets provide much higher risk-free adjusted returns when a democrat is in office.  The difference is large – 9% per year for a value-weighted portfolio.  For an equally-weighted portfolio, the difference is 16% (the difference for small-stock returns was larger than for large-stock returns).  The tests use data up to 1998 (the difference is likely larger now).

The results were robust to a battery of tests, over different time periods, and after adjusting for the business cycle.

The authors also found that this stronger performance during democratic years exists in the presence of less volatility.

This evidence flies in the face of the common belief that Republicans are more beneficial to the markets than Democrats.

Reference:

“The Presidential Puzzle:  Political Cycles and the Stock Market” By: Santa-Clara, Pedro; Valkanov, Rossen. Journal of Finance, Oct2003, Vol. 58 Issue 5, p1841, 32p

Analyses of HOLDRs

BBH:  Bullish.  (Biotechnology)

SwingTracker
MrSwings Real-Time Stock Charts RISK-FREE TRIAL featuring one-click access to Larry Swing's profit-generating indicators - Force Index, EquiVolume, True Strength Index

BDH:  Bullish on a breakout above Friday’s highs.  (Broadband)

BHH:  Neutral.  (Business to Business)

EKH:  Neutral.  (European Stocks)

HHH:  Neutral.  (Internet)

IAH:  Neutral.  (Internet Architecture)

IIH:   Neutral.  (Internet Infrastructure)

OIH:  Neutral.  (Oil Services)

PPH:  Neutral.  (Pharmaceuticals)

RKH:  Neutral.  (Regional Banks)

RTH:  Neutral.  (Retail)

SMH:  Neutral.  (Semiconductors)

SWH:  Neutral.  (Software)

TTH:  Neutral.  (Telecommunications)

UTH:  Neutral.  (Utilities)

WMH:  Neutral.  (Wireless)

Analyses of Individual Stocks

Bullish Stocks

IBM

• Consolidation breakout
• Unfilled gap
• Bullish volume