Algo traders: What risk management are you using to manage crisis like Flash Crash

[2rosy]

Well, I dunno if you witnessed the Flash Crash in person, as I did, there's NO WAY you would be able to turn on the "kill switch". The S&P tumbled over 100 points (can't remember the exact number, but huge amount for that time) in a matter of minutes--viz. blink of an eye.

i did and minutes are a long time which is why everyone turned off there algos asap.

 

[MacBookProHo]

Algo traders should know their stats on drawdown and should probably stop trading if hit

It all sounds like everyone is trying to invent a system that beats roulette,

dID Anyone crack the ATM money machine code Holy Grail yet ?,
That's not trading skill, process or understanding -- But rather, just, finding slight, tiny, imbalances to quickly exploit within the market system,

Ironically, the numbers on a roulette wheel adds up to 666....the Devil is toying with you all,

 

[ValeryN]

Business as usual in my case. Backtests over 70 years show the best action is to do nothing, assuming no doubling down/5-7% exposure to individual stock/limiting new positions taken.

Typically running 5 strategies. Some long some shorts. That helps thru flash crashes.

Only monthly evaluation of model vs live. Wouldn’t know if something happened in the market otherwise.

Interesting thread.

 

[guest_trader_1]

- Volatility indices and indicators to measure the market range.
- Set dead man switches accordingly. Halt trading on exceptional events.
- Always trade both sides of the book on separate accounts to offset loses.
- Never trade more than 30% of the buying power.

 

[blink18]

For those who automate, I'm just curious what risk metric you consider as sufficient to get around catastrophes like Flash Crash. Do you have a pulse on sudden volume spike to prevent you from entering? How would you manage to get out if you were already in a trade (although there's not much you can really do at this point)?

Indexing and diversification across different asset classes. Biggest losses were with individual stocks, I remember some stocks went down a lot (-50%,-70%) and some people sold/bought at those levels. Crash and rally back to even, happened in one hour.

 

[Lucas_Paul]

Well, I dunno if you witnessed the Flash Crash in person, as I did, there's NO WAY you would be able to turn on the "kill switch". The S&P tumbled over 100 points (can't remember the exact number, but huge amount for that time) in a matter of minutes--viz. blink of an eye.

a minute is forever in algo trading, so i'm not sure what you are asking here?

The market wasn't probably empty the whole 100pts, i'm sure you're algo would have gotten out with "decent" slippage so you take a little larger than normal loss ASSUMING you are on the wrong side AND flash crash happened RIGHT IN FRONT of your predefined stop. Otherwise you'd have room anyway till your stop and slippage wouldn't have been that large.

other than that there's no other way but to have options in place all the time and constantly adjusting them (having a strategy around them besides your regular market strategy. It's called cost of doing business

 

[Lucas_Paul]

- Volatility indices and indicators to measure the market range.

His talking about "force majeure" event, measuring historical volatility isn't helping here.

 

[guest_trader_1]

His talking about "force majeure" event, measuring historical volatility isn't helping here.

How do you know that is an exceptional event if you don't measure historical volatility? You know it because the market range is exceeding the expected market range levels.

So measuring historical volatility is key.

 

[Lucas_Paul]

How do you know that is an exceptional event if you don't measure historical volatility? You know it because the market range is exceeding the expected market range levels.

So measuring historical volatility is key.

He'a talking about a FLASH CRASH.

if your flash crash 5min bar high-low = 100pts+ while your avg 5m bar has it 5pts AND the market don't fill you, why are you measuring that historical volatility for?

 

[earth_imperator]

He'a talking about a FLASH CRASH.

if your flash crash 5min bar high-low = 100pts+ while your avg 5m bar has it 5pts AND the market don't fill you, why are you measuring that historical volatility for?

But it could fill, ie. data still could come, even amid a flash crash. I mean it could...
And normally such auto-trading uses shorter time frames or bar sizes, ideally streaming data.