Knowing whether the stock is likely to go up or down is dangerous without having some model of how much it will go up or down. It could have a 90% chance of going up but with a negative expected return (because if it goes up it will only go up slightly but if it goes down it will crash).
I wonder if this guy understands this? His "up/down" language suggests that maybe he doesnt and his group is only making money by "picking up pennies in front of a steamroller".
Monty Hall and the Birthday problem? Are they still asking these old hackneyed interview questions!?
As a whole everything this guy said in the interview sounds a bit naive/old-hat. Are people really still blindly trading the correlation between MSFT and Oil?
The whole interview reads like what a junior statistics graduate thinks quant trading funds do rather than how they actually make money nowadays.
It's like he hasn't even read Fooled by Randomness....
Companies will continue to use old and well worn questions, and variations on the themes, in order to test candidates statistical intuition. Note that the interview questions asked for every job are just as standard and game-able.
I was surprised he didn't even mention the kinds of programming languages that would be useful.
-- zero value added, in other words.
 Front running is not the only use of statistics and quants in finance, however.
"But," you may argue, "Siri changed from when Menlo Ventures invested in it to when Apple bought it."
The VCs also participated in an $8.5 million funding round that didn't exit until over 1.5 years later. Microsoft changes from quarter to quarter and, more importantly, the world economy, upon which Microsoft's prospects are fundamentally (and to a degree predictably) related, produces a machine gun chatter of data points throughout the day (with infinitely more interpolatable).
If we think of the stock market as a resource allocation decision engine that learns via Bayesian inference then individuals are constantly signalling to one another with information being absorbed first by risk-tolerant traders (analogous to early adopters) and progressively by more risk-averse ones.
Forecasting which assets investors (that, in aggregate, comprise a weighted sample of society) want to buy in the future boils down to figuring out what people will want to direct resources to in the future and then directing resources to those things today. Not that different from trying to stay on the bleeding edge of the tech curve, either.
TL;DR Calling forecasting-driven investing front-running dilutes the term by capturing such (clearly to HN) productive activities as tech.
Fortunately, you are not alone in this error . Unfortunately, it is still irrational.
EDIT (reply to below):
>However, logically, If there was value being added, this could be extracted absent harm to client. But since there is no fundamental value being added, this is impossible.<
This makes no sense. This almost seems like a tautology.
Are you going to say that everything in society with no added value (value being vaguely defined to the point of being meaningless) is (or should be) illegal?
No, It's more technically the inevitable harm it causes the client (which is a conflict of interest). However, logically, If there was value being added, this could be extracted absent harm to client. But since there is no fundamental value being added , this is impossible. And so at a logical level, its like statuatory rape. Its illegal because its deemed no good can come of it.
Just because something has no added value does not make it illegal.
True. But Nobody is arguing this. There is a correlation/causation fallacy to be mindful of. There is alot of overlap (correlation) but its not perfect (~causation). This does not rule out that where it <is> illegal (overlap) the cause is related to the value added/not issue.
[edits-- to respond/clarify per new comments above]
 Front running in the "true sense" is this: client wants to buy X amount of asset Z for at the market price. places order with dealer [with our without a limit P]. Dealer buys X (or a subset) of asset Z cheaper than P and sells at up to P, taking the difference as profit with zero net/residual risk in the underlying. If the dealer did not intervene, the client would be better off. The trade is only profitable for the dealer because it is risk-less, but it is risk-less only because he has <information> the client <gave> him. That info was "given" under the presumption that it would not be mis-used (ie, by front-running, etc). Since this info needs to be given (1) for <any> trades to happen; and (2) is inherently subject to this type of zero-sum abuse; it is historically unethical/illegal/etc for this type of front running to be persued by licensed BDs (or equivalents). This is distinct from market making (where risk is being taken), etc. and other legitimate forms or speculating, etc.
Equity markets have massive economic externalities. This is true dimensionalized by mean (valuation) of price levels and variance (of price changes). There is a strong public policy rationale that markets be fundamentally fair, information efficient, and also <accurate> in their moves (ie, such that vol is ~correlated to changes fundamentals, not 'animal spirits', etc). A market with prices subject to random error variables that distort variance (even if mean neutral) is fundamentally <worse> than one without such pricing characteristics, for example.
If it takes 30 pages of stochastic calculus to explain the utility provided by a fund, the public is going to dismiss the fund as "providing no value" and the fund isn't going to bother even trying to defend itself, because it just sounds like excuses, even if there is very real value being provided.
Many times the fallacy is in assuming that all market participants care only about average returns, when risk/volatility play a big role in the utility functions of the vast majority of market participants.
For a very simplified example, let's say QFundA has discovered some very reliable heuristic for the movements of a variety of small cap retail equities over the one-week timeframe. Let's also say that Pensions R Us has decided to sell 50% of the average daily volume of a regional sporting goods retailer over the next 2 days. If QFundA's model strongly suggests that the price is going up over the week, QFundA may slowly take on those shares over 2 days (and hedge against big news using out of the money puts), followed by slowly selling them off over the following 2 trading days to market participants who weren't interested a few days earlier. QFundA has made money by smoothing out prices (and likely reducing the bid-ask spread). Pensions R Us gets a better average price than they would have otherwise, as QFundA's buying activity has offset the market impact of Pensions R Us's selling activity.
Impatient/risk adverse market participants may have lower average returns because of QFundA's activities, but perhaps they're more than happy to pay a few basis points in order to get into or out of their positions faster. It's entirely possible that QFundA only slightly reduces average returns of other market participants while greatly reducing volatility, and thus bringing up everyone else's Sharpe ratios. In other words, when you take into account the fact that volatility/risk pays a big role in the utility functions of most market participants, it's possible that everyone wins, despite QFundA providing "zero value added" from the narrow viewpoint that only looks at the long-term price prediction provided by long-term "fundamentals" investors.
QFundA's strategy could still be summarized as "I know [predict high probability that] somebody else will buy X, so let's buy X first and sell it to them at a higher price", and yet QFundA provides real value to the market as a whole.
There are some predatory strategies that don't provide any value to other market participants. However, just because you don't understand the value provided by some market participant (and they know that trying to explain it is just going to sound like nonsense excuses to someone with your level of understanding) doesn't mean that no value is added.