FYI: QuantyCarlo released a free online historical option data viewer.
http://quantycarlo.com/announcing-quantycarlo-data-viewer/10386
http://quantycarlo.com/announcing-quantycarlo-data-viewer/10386
Free online historical option data viewer
- Thread starter paladincomp
- Start date
Do you mean you don't need intra-day data, or that 15 minutes is not sufficient granularity for your needs?
I think for covered calls, LEAP strategies, etc., EOD is perfectly adequate for backtesting. But for someone who trades weekly options, intraday data historical data is essential. Some would even argue that more granular than 15 minutes is desirable. Perhaps that's why such vendors as Market Data Express sell e.g. 5 minute and 1 minute increments.
I think for covered calls, LEAP strategies, etc., EOD is perfectly adequate for backtesting. But for someone who trades weekly options, intraday data historical data is essential. Some would even argue that more granular than 15 minutes is desirable. Perhaps that's why such vendors as Market Data Express sell e.g. 5 minute and 1 minute increments.
I would argue that 15 minute snapshot data are pretty useless in any intraday trading strategy or any strategy for that matter that has holding periods of up to several hours. Even 1 minute snapshot data are of very little use because in options trading, due to much less liquidity in most contracts a lot can happen in a minute to which most strategies could be very sensitive to but backtests, perusing such snapshot data, would not account for.
If you are getting in and out in the same day, I see your point. On the other hand, if your game is primarily theta or vega plays (symmetrical Iron Condors, Iron Butterflies, Calendars, Double Diagonals etc.) you need to stay in the market days rather than minutes, and touch the position as little as possible in the mean time. For such strategies, frequent adjustments and reacting to every little move of the market leads to death by a thousand cuts.
why would I need historical intraday option price data for the purpose you outlined above?
As I see it, the only reason would be to optimize entry rules, adjustment rules (balancing delta or vega, for example) or exit rules. Sure, you could do this with EOD data, but then you would be ignoring the intra-day market whipsaws, which could trigger adjustments or exits, but would not show up when only using EOD data. And again, especially if you trade weeklies, you definitely want to validate your algorithms on intra-day data. At the other extreme is the covered call writer, who can just "set it and forget it," and therefore would not need intra-day data.
well then again if you wanted to do that then 15 minutes snapshots nor 1 minute snapshots would be useful because prices in between such snapshot data might, as you yourself stated, "trigger adjustments or exits". I have a feeling we are getting into circular arguments but here the gist: Either you trade a multi-day/week/month strategy and only trade at market open or close for which intra-day historical data is unnecessary or you need precise intraday data and time stamps, but data in between is quite useless imho. I never understood why someone would show interest in intra-day snapshot data.
I remember doing a study on deltahedging SPX gamma during the overnight (this was like 8 years ago). It was far easier to run the study using intra-day snapshot data. I took Bloomberg intra day 5 minute increments and got the high, low, open, close for those 5 minutes. I would assume that if my hedging trigger happen within that range, I would get filled (generally reasonable for the market condition at the time). It showed me if there was enough volatility to justify hedging in the overnight vs waiting for a gap move on the open.
So it gave me the result I needed while being simpler to manage. I could have done it with precise tick data and perhaps gotten a slightly more accurate result but the snapshots were fine and at the time the precise data was a limitation of Bloomberg which meant a lot more energy sourcing and building infrastructure around a new dataset for a study that wasn't worth that much pnl.
So it gave me the result I needed while being simpler to manage. I could have done it with precise tick data and perhaps gotten a slightly more accurate result but the snapshots were fine and at the time the precise data was a limitation of Bloomberg which meant a lot more energy sourcing and building infrastructure around a new dataset for a study that wasn't worth that much pnl.
I appreciate having my assumptions challenged, because it forces me to think more deeply. I think part of the rationale for discrete time slices rather than tick data is convenience of being able to process things more quickly. But that does not tell us anything about the adequacy of 15 minute slices versus tick.
Let's suppose we can choose any arbitrary time sampling interval, and that our goal is to reduce the chances of a big price move going undetected, i.e. happening in-between two of our samples. Of course smaller is better. How much better?
Well, it turns out that we can answer this with simple statistics. Let's define a "big price move" as exceeding "a 1 day 1 standard deviation price move." Assuming a normal distribution, to keep things simple, the probability of this happening in a 24 hour period is 31.73% from the definition of normal distribution. Now let's say that that probability is too high for me, so I want to reduce it to a "reasonably small" probability by making the interval smaller.
Now, since there are only 6.5 hours of actual trading when all the price movement takes place, we scale our time axis by 6.5/24.
Well, as you can see in the attached graph, to get a very small probability for a 1 stdev price movement we need an interval size smaller than one hour. But 30 minutes gets us pretty close to 0%. If, on the other hand, our adjustments are only triggered with a (one day) 2 stdev price movement, a time interval of two hours is already quite close to 0%.
I think this would be the sort of logic one might use if arguing that 15 minute increments were "fit for purpose," with the emphasis on a clear definition of "purpose."
Let's suppose we can choose any arbitrary time sampling interval, and that our goal is to reduce the chances of a big price move going undetected, i.e. happening in-between two of our samples. Of course smaller is better. How much better?
Well, it turns out that we can answer this with simple statistics. Let's define a "big price move" as exceeding "a 1 day 1 standard deviation price move." Assuming a normal distribution, to keep things simple, the probability of this happening in a 24 hour period is 31.73% from the definition of normal distribution. Now let's say that that probability is too high for me, so I want to reduce it to a "reasonably small" probability by making the interval smaller.
Now, since there are only 6.5 hours of actual trading when all the price movement takes place, we scale our time axis by 6.5/24.
Well, as you can see in the attached graph, to get a very small probability for a 1 stdev price movement we need an interval size smaller than one hour. But 30 minutes gets us pretty close to 0%. If, on the other hand, our adjustments are only triggered with a (one day) 2 stdev price movement, a time interval of two hours is already quite close to 0%.
I think this would be the sort of logic one might use if arguing that 15 minute increments were "fit for purpose," with the emphasis on a clear definition of "purpose."