- ago
Hi there,

when watching the tutorial videos, the backtesting is usually done with market index datasets or even the survivor bias-free datasets (which I don't know where to find it btw :) but ... is anything wrong on using my own list of carefully selected stocks (based on fundamental and technical analysis) that I trade manually nowadays? Of course, the list would have probably looked differently a few years ago and will also change in the future but I'd say such setup has the potential to beat the market. With my modified Knife Juggler strategy and backtesting in 5 years back, the APR is roughly 100% versus some 20% on QQQ.
Well, both seem to be a bit unrealistic to me but ... what is a realistic expectation? What are the typical profits one can expect here?

thx
Jiri
0
391
10 Replies

Reply

Bookmark

Sort
Cone8
 ( 23.73% )
- ago
#1
Worth a read if you haven't read it yet -
https://www.wealth-lab.com/blog/survivorship-bias

Worth a view if you haven't seen it yet -
https://youtu.be/CCPKgue5Ozg

Keeping the pitfalls in mind, sure, you can make your own DataSet. Targeting ETFs also works to remove survivorship bias.. and QQQ has a long-enough history to test though bull and bear markets.
1
- ago
#2
Sure, I did read and watch these but this does not answer my question. Having my strategy tested on a survivor bias-free dataset basically assumes that I'm going to trade it blindly on the index stocks. In that case, yes, this is perhaps the best approach to increase the probability that the strategy will be still profitable even with some stocks failing and others coming up over the months and years.
But this is not my intention.

And one more question actually... in the backtesting, I can compare the strategy against a specific symbol's buy&hold. Is there any way to compare against a list of symbols buy & hold? Basically, when backtesting the strategy on my own list of symbols, I'd like to compare the performance with the same list's buy&hold to check if the trading exercise makes any sense at all :)
0
- ago
#3
QUOTE:
Is there any way to compare against a list of symbols buy & hold?

I can imagine a way but it would be too convoluted to discuss it here: CompIndex indicator (using Index-Lab), export to .QX binary file, inject it into a DataSet of any compatible data provider, specify the symbol as your benchmark symbol.
0
Cone8
 ( 23.73% )
- ago
#4
QUOTE:
assumes that I'm going to trade it blindly on the index stocks
What's "blind" about that? I'd argue that blindness is testing any list of stocks that you invent "today".

I though I answered by indicating that if your trading target is only an ETF (or a group of them), these automatically account for survivorship.

QUOTE:
Is there any way to compare against a list of symbols buy & hold?
Just run a Buy and Hold strategy on that DataSet with Equal Shares sizing. To compare results for 2 tests side-by-side, just right click, copy, and paste the results to Excel.

Here's the Buy & Hold Strategy:


And here's the sizing. Make sure to add a little margin (1.1 shown) to account for gaps so you don't get an NSF.
0
- ago
#5

QUOTE:
Of course, the list would have probably looked differently a few years ago and will also change in the future but I'd say such setup has the potential to beat the market.


This is selection bias at its purest.

If you can define your criteria for your "list" selection you can also backtest that.

Best regards,
Richard Dale
Chief Information Officer
https://norgatedata.com/
0
- ago
#6
All my trading datasets are cherry picked (i.e. curated) based on each stock having a recently positively sloped equity curve with that particular strategy. Why would any trader want to include stocks in his datasets that didn't make money over time?
0
- ago
#7
Because the future is unknown and these stocks provide the necessary level of diversification in the changing environment (i.e. yesterdays' dogs becoming stars)?
0
ww58
- ago
#8
Counterweight to survivor bias is stock behavior. This means that a strategy that works on nasdaq100 may not work at all on sp100. Therefore, for me, custom dataset is not a problem, especially for smaller timeframe strategy.
0
- ago
#9
QUOTE:
... yesterday's dogs becoming [tomorrow's] stars ...
That's an interesting point. When I remove a stock from an actively trading dataset, I place it in an inactive dataset for future scrutiny. Many small and micro cap stocks have not recovered yet. Will they later? I don't know; some won't.

The other reason for the inactive dataset is for future refinement of a production strategy. Perhaps the strategy is too easily fooled by doggie stocks and needs additional metrics to sharpen its Buying choices.
0
- ago
#10
QUOTE:
If you can define your criteria for your "list" selection you can also backtest that.


That would be great but I'm not sure this is possible here.
I do my stock evaluation in MarketSmith and use multiple criteria like
- growing YoY earnings (if possible)
- growing YoY revenue
- growing institutional ownership (possibly by good-quality funds)
- Industry group ranking
- analysts estimate changes
- proprietary IBD metrics
That's on the fundamentals side
On the technical side, I try to make sure that the
- stock's relative strength (not RSI!) compared to the market is high
- the stock is in a strong uptrend
- ideally trending up in a channel
- ideally bouncing off the bottom edge of the channel before the purchase
0

Reply

Bookmark

Sort