|
PureBytes Links
Trading Reference Links
|
There's another problem:
In the literature and on the 'net you'll find a lot of the time that people
compute the Sharpe ratio for a fraction of a year (say a month) AND THEN
ANNUALIZE IT. Which is of course complete nonsense, but it explains why
the numbers bandied around are so high.
William F. Sharpe kindly put the proper way of doing it on the 'net - just
google for the name.
Regards,
Stefan Schulz
At 18:48 27/03/02 -0500, Michael de la Maza wrote:
>I am constantly astonished by the "requirements" that retail traders
>need a mechanical trading system to meet before they are willing to
>trade it.
>
>An excellent example is below. No mutual fund, hedge fund, commodity
>futures program, etc. in the history of the world has ever had a Sharpe
>ratio of 4 for any extended period of time (and an annualized return in
>excess of 15%).
>
>John Henry, member of the Forbes 400 and owner of the Boston Red Sox,
>has personally made over $1 billion in the past twenty years trading a
>commodities portfolio that has a Sharpe ratio of about 0.4. He loses
>money five out of every twelve months and regularly has drawdowns in
>excess of 20%.
>
>John Dunn has a twenty five year track record and manages over $500M.
>He has a Sharpe ratio of about 0.5 and has had several drawdowns greater
>than 30%.
>
>Warren Buffett, George Soros, and Julian Robertson (all members of the
>Forbes 400 who made their money trading/investing) have lifetimes Sharpe
>ratios well below 2.
>
>But to listen to many retail traders speak, they would be unwilling to
>trade the strategies that John Henry, John Dunn, Buffett, Soros, and
>Robertson have used to make billions because these strategies don't meet
>their "requirements." That's unbelievable.
>
>In my mind, there is no doubt that the greatest difference between
>professional traders and retail traders is that professional traders can
>make tens of millions of dollars from strategies that retail traders
>consider to be untradable. The difference is not commissions or capital
>or money management.
>
>Traders who are able to move past this obstacle do very, very well.
>Examples that we all know include Mark "Aberration" Johnson and Mark
>"Oddball" Brown.
>
>Michael
>
>-----Original Message-----
>From: John Hayden [mailto:jhayden@xxxxxxx]
>Sent: Wednesday, March 27, 2002 9:04 AM
>To: omega-list@xxxxxxxxxx
>Subject: Criteria for determing a "good" trading system Part 2
>
>part 2 continues
>---------------------
>
>Once we know the amount of heat we can endure we must decide on how we
>will
>measure the performance of the trading logic. Below is a list of the
>common, and not so common indicators. However, what determines what a
>"good" value is for these indicators will depend largely upon the
>traders
>answers to the above questions. Here is a summary list. The minimum
>performance levels are based on trading a 1 lot S&P on a 3 minute chart,
>
>and doesn't account for slippage or commissions (as this seems to be the
>
>standard practice):
>
>1. Net Profit per Trade (Avg. $ per Trade) $ 200 minimum
>2. Profit Factor (Gross Profit/Gross Loss) 2.0 minimum
>3. Drawdown/Run-Up Analysis
>4. Sharpe Ratio 4.0
>minimum
>5. Total Trades 2-6/day
>max
>6. Total number of Longs/Shorts Ratio ~1:1
>7. Total Net Profit n.a. -
>subjective
>8. Average Winning Trade
>9. Average Losing Trade
>10. Average Win To Average Loss Ratio Depends on %
>winners
>11. Largest Loosing Trade/Average Win Ratio 1.5 or lower
>12. Largest Drawdown/Average Win Ratio 7.0 maximum
>13. Average Drawdown/Average Win 1.0
>Maximum
>14. Percent of Winning Trades n.a. -
>subjective
>15. Max. Consecutive Winning, & Loosing Trades n.a.
>16. Average Number of Winning, and Loosing Trades
>
>Gross Profit per Trade (GPT)
>
> Calculation: Total Profits divided by Total Trades
>
>This number is perhaps the most important because it tells us the
>average
>profit we can expect to make on each trade. Each trade will generate two
>
>associated cost; slippage, and commissions. Should the GPT be low then
>slippage will change a profitable system to a loser. The higher this
>number
>the better. It is important to realize that when commissions AND
>Slippage
>are factored in many systems fall apart because this number is not
>higher
>than actual cost. Generally this number should be twice the expected
>amount
>of slippage - plus the commissions cost - for the S&P $200.
>
>Profit Factor
>
> Calculation: Total Profits divided by Total Loss
>
>This number will tell us the amount of profit we gain for every dollar
>placed at risk. Generally we want to risk $1 for the possibility to gain
>at
>least $2 dollars.
>
>Drawdown Analysis - Consecutive Losing Periods
>
>1. Determine the percentage of money lost (maximum)
>2. Identify the length that it took to hit this maximum percentage loss.
>3. Identify how long it took for the Recovery to take place.
>4. The Peak is the time frame just before the drawdown began, or the
>Start
>Date (which is one time frame less).
>5. The Valley is the time frame that the worst percentage loss was
>experienced.
>6. Identify the maximum amount of money lost.
>7. If not using time to identify length of drawdown then determine the
>number of trades before the equity is restored.
>
>Run-Up Analysis - Consecutive Winning Periods
>
>1. Determine the maximum percentage of money made in the Run-up.
>2. Identify the Start date the winning streak began.
>3. Identify the End date of the winning streak.
>4. Identify the length of time that it took to hit this maximum
>percentage
>gain.
>5. Identify the maximum amount of money gained.
>6. If not using time to identify length of run-up then determine the
>number
>of trades before the equity is begins to decrease.
>
>Generally we want to avoid any systems that experience more than a 30%
>loss
>of capital. We must totally avoid systems that generate 50% or more of a
>
>drawdown as it becomes almost impossible to recover from such a large
>reduction in ones equity. In judging the length of a drawdown to endure
>it
>is generally best to look for a system where the total length of the
>longest drawdown is no more that twice as long as the total length of
>the
>largest run-up. This is a subjective rule and is totally dependent upon
>the
>trader and the amount of pain he can take.
>
>Sharpe Ratio
>
>The ratio is based upon comparing the standard deviation of the system
>returns and the amount of risk of that system - to the amount of return
>that could have been earned in a risk free investment. Generally the
>higher
>the Sharpe Ratio the better the system. The Sharpe Ratio is a good way
>to
>compare different strategies as it levels the playing field. Again this
>is
>a subjective evaluation however generally I would like to see a value of
>4
>or higher.
>
>Total Trades
>
>The concept is simple; can the trader actually trade the system? How
>many
>trades a day (or whatever unit of time the trader wants to use) is the
>trader comfortable with? How easy is it to execute the trades? Is the
>execution method reliable? Generally I like 4-6 trades per day, however
>this is dependent upon the trader.
>
>Total Gross Profit
>
> Calculation: Total Profit less Total Loss
>
>This number is totally dependent upon the trader. Largely depends upon
>if
>the trader is trading full or part time. Must he trade to pay the bills?
>If
>so, then the total net profit required will be dependent upon the
>traders'
>lifestyle. This doesn't include commissions or slippage.
>
>Average Winning Trade
>
> Calculation: Total Gross Profit divided by total winning
>trades.
>
>This number tells us the average amount of profits per winning trade.
>
>Average Loosing Trade
>
> Calculation: Total Gross Loss divided by total loosing trades.
>
>This number tells us the average amount of loss per loosing trade
>
>Average Win To Average Loss Ratio
>
>Calculation: (Total Profits divided total number of winning trades)
>divided
>by (Total Loss divided by total number of losing trades).
>
>This ratio is based upon the average gain of winning trades, and the
>average loss on the loosing trades. While it seems similar to the NPT or
>
>Profit Factor it is not. This is because the other primary variable is
>the
>percentage of winning trades. With this ratio we can have a profitable
>winning system that will generate larger loosing trades than winning
>trades. The reason this is profitable is because the percentage of
>winning
>trades is more than 50%. Generally I want to see at least 2.5 or more -
>250% more average gain than loss.
>
>Largest Loosing Trade/Average Win Ratio
>
> Calculation: Largest Loss divided by Average Profit per winning
>trade.
>
>This ratio will indicate approximately how many winning trades it will
>take
>to recover from the largest losing trade. Generally a value of 4 or
>less
>is good, I prefer 1.5 or better. As this is telling us that with 4 (or
>1.5)
>average winning trades we will recover from the largest historical loss.
>
>Again this relates to the amount of pain a trader can endure.
>
>Largest Drawdown/Average Win Ratio
>
>Calculation: Determine amount money lost in the largest Drawdown divided
>by
>the Average Profit per winning trade.
>
>This ratio will also give us an indication of the amount of pain that
>must
>be endured. We will know the approximate number of winning trades it
>will
>take to recover from the largest historical drawdown. When combined with
>
>percentage of winning trades we can then begin to estimate how many
>trades
>overall must be traded. Generally I do not want to see a ratio above
>7.0.
>In other words this is telling me that in 7 winning trades I should
>recover
>from my drawdown.
>
>
>Average Drawdown/Average Win Ratio
>
>Calculation: Determine the average amount money lost in the all of the
>Drawdown's divided by the Average Profit per winning trade.
>
>Similar to the previous ratio we will know the approximate number of
>winning trades it will take to recover our lost equity. When combined
>with
>percentage of winning trades we can then begin to estimate how many
>trades
>overall must be traded. Generally I want the value to be no higher than
>1.0.
>
>Percent Profitable
>
> Calculation: Total number of trades divided by total number of
>winning trades.
>
>This ratio relates to the number you times you can make a losing trade
>without getting upset or doubting the integrity of trading logic. In my
>opinion it is the least important of the ratios in determining the
>overall
>profitability of a trading system. Generally I would like to see 50% or
>more - however it really doesn't matter all that much to me. The
>percentage
>of winners is totally dependent upon the traders' psychological makeup.
>
>Maximum Number of Consecutive Winning and Loosing Trades
>
>Calculation: Determine the maximum number of consecutive winning and
>loosing trades.
>
>Average of Maximum Number of Consecutive Winning and Loosing Trades
>
>Calculation: Average the maximum number of consecutive winning trades.
>Average the maximum number of consecutive loosing trades.
>
>These two numbers will begin to tell us what we could expect in actual
>trading and will be one of the first indications that the trading logic
>has
>gone astray when trading in real time. They are also totally dependent
>upon
>the psychological makeup of the trader.
>
>A word of warning.
>
>The challenge of creating a trading system for an experienced trader is
>that he is attempting to create a mechanical version of himself.
>Computers
>are incredibly dumb machines and cannot change their "programming" on
>the
>fly. They are very systematic and they are not discretionary. This is
>their
>weakness and their strength. Most experienced traders are to a certain
>degree discretionary in their day-to-day execution of trades.
>
>Because computers are very systematic they will always execute the code
>in
>the same manner with the same results provided the inputs are the same.
>Should the inputs change then the output will change. The goal of every
>
>trading program is to execute the same buy regardless of when the data
>stream was started. In other words if we are using 3-minute historical
>data
>that starts on 4/1/01 and ends on 4/1/02, and our logic only needs 30
>days
>of 3-minute data. If our system generates a buy on 10/1/01 1:30pm, and
>then
>exits the trade at 10/1/01 2:15pm our goal is for the buy to be
>generated
>for 10/1/02 1:30pm regardless if the data stream being used starts on
>4/1/01, or 8/1/01.
>
>All too often this is not case, the date and time the data stream begins
>
>will prevent the trade on 10/1/01 at 1:30 from being triggered. The goal
>is
>that the trade consistently be triggered.
>
>Another problem is that using historical data to create a trading system
>
>will in actuality create a system that has been "curve fitted" to some
>extent - regardless if optimization studies were performed on the
>variables. This is because the system creator has the vision of 20-20
>hindsight. Curve fitting becomes apparent data from a different time
>period
>is used.
>
>This is why the adage you read everywhere "past performance is not
>indicative of future performance" - because it is the truth. This is
>true
>of a computer based trading system or an individual trader!
|