Hacker News Reader

Comment by roymurdock | Hacker News Reader

> The number of assets registering large moves—four or more standard deviations away from their normal trading range—has been increasing. Such moves would normally be expected to happen once every 62 years.

This is very troubling. It means traditional models are not working to capture the market correctly at the moment. It was a leading indicator for the 2007 financial crisis - 3 sigma events happening with increasing frequency that were ignored because there were profits to be made from poor risk management.

View on HN

The 62 year estimate is based upon an assumption of normally distributed data. Everyone in the industry knows the distributions are significantly fat-tailed. Maybe things are getting more volatile, but the 62 year estimate is just mumbo jumbo.

Edit: also implicit in the 62 year estimate is constant volatility, but the same article mentions changes in the volatility of volatility... third order changes in something he was assuming was constant earlier.

Totally on point. That is why 100-yr events seem to happen every 7 or 8 years :-) Normal distribution assumptions are great for coursework, terrible when you are managing real money.

Slightly off-topic but we are really bad at predicting these "very unlikely events" that actually happen more often than expected. Whenever I hear a similar story, I am reminded of the 3 nuclear reactor disasters (Three Mile Island, Chernobyl, Fukushima), each of which shouldn't have happened in 15000 years.

Where did you get the 15,000 years number? I have never seen that before and would love it if you could provide more resources.

It is very anecdotal, I believe I read it in an article long time ago around the time the Fukushima disaster happened. It is probably an inaccurate number.

The point is, all 3 faults/human errors shouldn't have happened according to statisticians/mathematicians.

Maybe there could be some technical result like "if everyone in the market assumes distribution X and trades accordingly, that makes the real distribution more fat-tailed than X"? No idea if something like that could be true...

Or maybe human behavior and personality consistently resists the enforcment of technical rules, and watches for, and then willfully denies attempts to restrain behavior it (subsets of the collective population) instinctively desires to engage in.

Maybe some of us know that we're being watched, and simply defy any stupid egghead's function curves, for the sheer amusement of seeing disappointed faces.

Maybe there are sadistic people roaming free, and they feed on the souls of the damned.

How would that fit in with modern quantitative economics?

If some quants can predict these alleged flights of fancy, they'll trade against them and make some good money.

Acting, or according to your context merely "betting" against an adversary is fine, but this assumes the capacity to act against an adversary, and that the actions taken will be big enough to have an effect.

You can't realistically take actions against a belligerent force that has an overwhelming capacity to watch your every action and imprison or execute you. You can't take actions against forces that exist outside a legal framework and the juridictions you are subject to.

Placing bets is not taking direct action. It means sitting on the sidelines, and hoping that legal mechanisms take their natural course in a civilized society.

Well, that would depend on the size of that population, would it not?

A single middle-class person working to make as contrary to prediction market choices as possible would, I think, have a not significant influence

Even if they optimized for the size of the influence as well as the strangeness.

I don't expect the total population of people doing that to have a significant impact on the market.

There isn't much reason to do it, and it probably costs money over time, leading the people who do that to have less money over time,and so for their doing that to have less and less of an impact over time.

Not all people carry the same degree of influence over a market.

Not all people act according to or within the rules of a market.

Not all human behavior can be scoped to a form of market activity.

What would the bald destruction of a market, or set of markets, represent to the context of such narrow minded economics?

A normal distribution assumes independent random samples. That clearly doesn't happen in a market where people and algos are reacting to the same news. Moves end up more like step functions in both panicked or excited market and will then overshoot buying and selling.

I thought that it was customary to add a 'smile' to most financial distributions (i.e. increased probability of extreme events), I don't think that anyone believes that a normal distribution accurately measures anything much in finance.

So the real question is, if you think the market is gunna go down in flames soon, how do you short... everything?

You buy an ETF that shorts the S&P500: http://www.proshares.com/funds/sh.html

Is there a way to bet that something will grow more slowly than normal? S&p 500 has been growing about 10% per year. How would I "short" it and bet that it will grow 5%? Can such a position be constructed?

Straddle option on SPY with K=1.05 x SPY today.

https://en.wikipedia.org/wiki/Straddle

If you want to get clever and derisk a bit you can also limit your losses in the tails with way out of the money calls and puts.

If you have never traded before, don't try to jump in without knowing all the possibilities. A more likely scenario to unfold, is that Janet Yellen does NOT raise interest rates and instead, starts QE4. In this scenario, the dollar will tank and stocks (priced in dollars) will go up. Just be cautious.

Be careful with inverse ETFs especially if they are leveraged to move 2x or 3x in the opposite direction of the underlying index. They try to be inverse for each single trading DAY so if the market moves back and forth you can take geometric losses. For instance, if you have 100 dollars in an inverse etf and the market goes up 10% your inverse etf will go down to 90 dollars. The next day the market drops 10% so you think "hey my inverse ETF will go up 10% so I'm back to where I started" Nope a 10% gain on 90 dollars puts you at 99 dollars. If the market moves sideways for a while your investment will continuously decline.

So many ways... For small timers: inverse ETFs, put options, CFDs (for non-US folks) For big timers: the above plus: short futures, OTC options, equity swaps, CDS, plus anything bespoke you can afford

I haven't looked into it in years, but my recollection is that you don't buy the inverse ETF, you short the corresponding long ETF.

That's because there is inherent "decay" in the price of the ETF - if the market just keeps going sideways, the ETF slowly trends downward over time.

So if you buy the inverse ETF and the market doesn't go up at all, you still don't have a profitable trade - you should short the long ETF.

Is that still the case ?

Or, just buy the ETF that preforms well when there is an economic downturn. Possibly GDX and GDXJ (gold mining ETFs).

No, you're missing the point. If my understanding is still current (I suspect it is) these ETFs "decay" over time.

Which means you can be "right" but if what you're right about doesn't happen fast enough, you still lose money.

That's why people suggest shorting the ETF that is opposite your prediction.

Wait for it to tank, then buy.

Catching a falling knife is always a good idea.

High reward goes with high risk. I honestly think there's a gap in the market for a falling knife fund.

Whereas catching knives thrown on upward trajectories...

Would happen if you were shorting

When you catch all the falling knives with an index fund, some will sink, but most will rebound.

There's Taleb's "tail risk fund".

Look at Exter's pyramid and decide which level(s) can withstand whatever you flames you think are coming, and then divide your assets among those levels in proportion with how bad you think things might get.

https://en.wikipedia.org/wiki/John_Exter

Central planners delegating to econometric models that are based on statistical historical patterns wonder why the "system" doesn't do what they expect; either its time for some new theories, or rediscover some economic schools of thought that have consistently rejected the historical schools.

How do you know options skews aren't already pricing this factor in? Usually this effect is counterbalanced by the vol smile. Edit: vol smile is how options very far away from the strike tend to be priced at a premium versus what black scholes would suggest. The sellers of these options demand a higher premium than "normal" to account for the fact that the data isn't actually normally distributed.

Reply on news.ycombinator.com