How to research and verify trading ideas

[talontrading]

Thank you. Solid work and hopefully enough to inspire some more interest in this little system!

your signals of course are slightly different than what i see on my screen because your criteria are slightly different. it would be interesting to see if 3 red candles works better than 3 closes down... it may.

good start. thanks.

Quote from JScott:

Attached is the low-budget quantitative analysis of this strategy going back three years. I used the DIA this time as a proxy for the DJIA – which had a few more signals. No fancy performance strategy, just a quick check of entry, exits and heat (risk). No stop baked in a la Talon, but a big stop isn’t needed based on this limited data sample.

Note: Entry signal was only after close was “less than” day before for three days in a row . . . not exactly the same as “three red bars” like I said before. I’d say that a little flexibility with the definition of “consecutive close down” might show many more, very good signals.

Also, percentile of 5 day close is often = 50%, but exit was determined when it exceeded 50%.

I like this strategy because it is surely very uncomfortable for nearly all traders to buy/sell into the face of momentum. Wouldn’t bother me if proper homework was done.

See attached for a summary of manual backtesting if you like. Thanks.

Keep trading.

J.Scott

 

[talontrading]

yes... you're buying weak stocks here... making sure that those stocks are not right on their long term lows is fairly important. i'm not saying you can't have a disaster buying a stock right on their highs, but historically most meltdowns come when stocks are closer to their lows.

this is the (misguided in my opinion) appeal of strategies that for instance only buy above the 200 day moving average. in backtests they would have avoided most of the really bad days.

your testing software will give an incorrect assessment of the risk in this strategy unless it includes a large selection of companies that have gone bankrupt or delisted for other reasons. our database does, but only because we built it an maintain it by hand.

Quote from 4XQs:

...increase risk?
...increase volatility of your returns?
...mandate lower position size?

 

[The_End]

I'm still a little confused. Below are the closing prices of AAPL for the last 15 days, from highest to lowest, according to YAHOO! Finance.

Could someone please explain to me specifically how the 10th percentile of the past 15 days was calculated to come up with 200.16? TIA.


207.00
206.63
205.96
205.88
204.45
204.44
204.19
203.25
202.98
201.99
201.46
200.59
200.51
199.92
194.34

 

[xtrader99]

Quote from The_End:

I'm still a little confused. Below are the closing prices of AAPL for the last 15 days, from highest to lowest, according to YAHOO! Finance.

Could someone please explain to me specifically how the 10th percentile of the past 15 days was calculated to come up with 200.16? TIA.


207.00
206.63
205.96
205.88
204.45
204.44
204.19
203.25
202.98
201.99
201.46
200.59
200.51
199.92
194.34

Use Excel formula =PERCENTILE(A2:A16,0.1)

 

[talontrading]

Apologies. The confusion here comes from Excel which uses a fairly non-standard calculation for percentiles. In our programming, we use R or C and only use Excel for quick and dirty analysis. Since I was at home, I just blasted Yahoo data into Excel and didn't think about the details.

Without going into a math lesson, Excel isn't exactly wrong per se, but it's non standard. Using Excel for statistical analysis is not a good idea, but we all do it sometimes because it is so easy to use.

Attached is a percentile output list from a real statistical analysis package... I think this will make more sense.

Again, apologies for the confusion. In this particular case the difference wouldn't have actually mattered because the close just needs to be in the bottom 2 of the last 15 days. If we change the 15 to another number, it might be more significant.

Calculating a percentile is about as elementary a statistical function as there is... not that much harder than simply counting... so the fact there can be a discrepancy here should remind us all to be on guard for all the little details. This is why we do everything in R or C or VBA and don't trust anything done, for instance in TradeStation or Wealth Lab.

AAPL
Percentiles
1% 194.34
5% 194.34
10% 199.92
25% 200.59
50% 203.25
75% 205.88
90% 206.63
95% 207.00
99% 207.00

Quote from The_End:

I'm still a little confused. Below are the closing prices of AAPL for the last 15 days, from highest to lowest, according to YAHOO! Finance.

Could someone please explain to me specifically how the 10th percentile of the past 15 days was calculated to come up with 200.16? TIA.


207.00
206.63
205.96
205.88
204.45
204.44
204.19
203.25
202.98
201.99
201.46
200.59
200.51
199.92
194.34

 

[The_End]

That clears it up and I understand now, many thanks to both of you for the help.

 

[JScott]

Quote from The_End:

That clears it up and I understand now, many thanks to both of you for the help.

For the quisitive types . . .

The attached explains why Excel (and maybe other platforms such as Tradestation) can have "unique" calcs . . . see pp. 5 - 6. Has to do with repeated values in a data set. Over a large data sample, probably near zero effect.

Using the function PercentRank in Excel probably is more user-friendly for this type of analyis (which is what I used in TradeStation).

Keep trading.

J.Scott

 

[xburbx]

i may have missed something in the last few posts on this system, but isnt this essentially the same thing as setting up something as simple as bollinger bands on a large timeframe? BB's show when price is inside/outside 95% of the price movement for that period? granted it doesnt address trade management or an entry method. the same is true for market profile as a concept.

 

[talontrading]

No it's not at all the same. Bollingers have a fatal flaw in their concept and we did discuss my bias against any indicator that is simply a derivative of price. With a few exceptions (Bollingers are not one of the exceptions) they just don't test out to have a quantifiable edge... another case of what the pubic thinks works, doesn't.

I also take it by your "something as simple as" phrase that you think this is more complicated than Bollingers. It is MUCH more simple... you only think Bollingers are simple because you click the button on your charting package and they magically appear on the screen, but in concept this is 1) much more simple, 2) is conceptually correct and 3) works.

Quote from xburbx:

i may have missed something in the last few posts on this system, but isnt this essentially the same thing as setting up something as simple as bollinger bands on a large timeframe? BB's show when price is inside/outside 95% of the price movement for that period? granted it doesnt address trade management or an entry method. the same is true for market profile as a concept.

 

[xburbx]

ok. sounds good. at what point would this strategy be deemed not good any more? if price breaks and holds in a new range and the percentiles are altered because it holds in the new range, what room for loss is acceptable? at some point there are new percentiles to deal with correct?