sp500 statistics since 1984

the main hypothesis is that there is an amount of tradeable distortion within the daily data of the sp future. toby crabel's work is well documented and I wondered about the narrow range day and its forecast quality in terms of trendiness. I failed so far in proving that it exists. An insidebar with lower range then than the average of the previous bars is not followed by a more trendy day than usual IMO.

Starting from there I began to analyze other things of that kind. The seasonalities were tempting yet seem to be too inconsistent.

I currently work on volatility and it's influence on daily gain.

I did some analysis as well on the relation between high/open and low/open and found that, although the market on average closes 0.04% higher than it opened, that it trades on average more down than up, which is surprising, yet was not significant enough to trade it.

I am interesting in cycle length as well, defined by the length between two slope changes on a moving average. It seems as if cycle lenght correlated negatively with volatility - not really surprising. I am very early stage here.

Finally I am patternising bars by giving the quintile for each OHLC within the range of recent x days. Found some mean reversion effect on weekly data, which I am paper trading now.

I do not expect any specific number to work perfectly, just try to build a set of effects that can be combined efficiently.


peace
 
man,

I'm not that familiar with Crabel's work, but I had a little different impression of the narrow range day(NRD) concept. I thought the idea was to go with a breakout of the NRD, not that the NRD itself was indicative of an impending move. In effect, the NRD represents a coiled spring that, if triggered, should lead to a good move.
 
I havn't read his book either. I just found in another thread the claim that we should have a trending day yesterday, since we saw a narrow range inside bar day on friday.

I understood the concept in the way that you better trade on breakout=trendfollowing after a NRD than bet on reversion games.


peace
 
absolute daily gain = abs(ln(Cto/Oto)) in dependence of recently observed daily range. absolute daily gain could be seen as the best proxy for trendiness observable on daily data.

environment
days for range 1
upper 50%
lower 0.00%
median daily gain 0.51%
mean daily gain 0.72%
observations 4835

reading
if yesterdays range (days for range =1) is below upper (50%) and above lower (0.00%) border then the median (absolute) daily gain is 0.51% with an average of 0.72%. since this wide range covers all cases this is all days since 1.1.1984.
 
environment
days for range 1
upper 50%
lower 1.00%
median daily gain 0.58%
mean daily gain 0.81%
observations 3003

environment
days for range 1
upper 50%
lower 1.50%
median daily gain 0.65%
mean daily gain 0.92%
observations 1545

environment
days for range 1
upper 50%
lower 2.00%
median daily gain 0.73%
mean daily gain 1.08%
observations 771

environment
days for range 1
upper 50%
lower 2.50%
median daily gain 0.89%
mean daily gain 1.36%
observations 383

environment
days for range 1
upper 50%
lower 3.00%
median daily gain 1.04%
mean daily gain 1.62%
observations 202
 
varying the lower border requirement for yesterday's range I see significant increase in today's expected absolute gain.
 
now I am looking for the relation between high and close within the daily range. if (H-C)/DR is high we are basically closing near the low of the day.
In addition I put now the change from closing to opening into it since it might be interesting in this context. Starting again "blank".

observations 4824
(H-C)/DR
upper 100%
lower 0%
range
upper 100%
lower 0%
Cto/Oto
median 0.04%
avg 0.02%
Oto/cT-1
median 0.01%
avg 0.01%
 
measure of trendiness for the day is ABS(O-C). Daily change is flawed because sometimes gaps account for 90% of the daily change.
 
traderkay
right on the gap thing. That's why I define dailyGain as ln(Cto/Ot-1) and yield as ln(Cto/Ct-1).


peace
 
You must log in or register to reply here.
Back
Top