Hacker News Reader

Comment by ddeck | Hacker News Reader

I've had colleagues that disconnected their juniors' mice to ensure this.

The reliance on excel in all aspects of investment banking is remarkable. From the simple stuff in corporate finance (merger analysis, capital issuance etc.) to complex derivatives pricing on the market side.

On the pricing side, external libraries typically handle the heavy lifting, but excel is the glue that brings it together. Certainly not the most efficient tool - I've had spreadsheets that take over an hour to calculate. You'd hit F9 and head out to lunch.

View on HN

Exactly - its efficiency is that the analysts can write the formulas themselves instead of briefing a development team and waiting a week.

Good point. The upside of Excel is that in return for predictably slow execution, you get a tool that can be used on the fly without learning SQL or adding a DBA layer to every task.

If you and your parent poster are right about why Excel is so prominent, and I personaly think both of you are, then we might be able to see a shift in economics and businessmanship and in the way Excel formulas play a large part of what it is that makes the world go 'round, once business men start to learn how to program, or when Excel start's to learn to talk "business", but no earlier than that.

Which movement do you think is strongest? The AI camp or the ones that think programming should absolutely be a part of a proper education in business or economics? I hope for the latter, because I think it will be the end of the huge gap there is today, between the decision makers and the engineers, two groups who quite often but for no good reason feel alienated from each other.

The division of labor here is due to time constraints, not a failure to understand value. A lawyer can also be a doctor, if not for the fact that they'd get little of either done.

If you make programs that 'speak business', then you make programs that are significantly more complex than those that 'speak computer'. Learning how to handle this complexity and optimize this kind of code is not something a business/economics person wants to spend their time doing. So they leave it to someone who specializes in it.

Yes, I see what you mean. But surely, just adding Programming to the curriculum of businessmen, who I'm sure are quite intelligent or else they should be considering other occupations, does not equate to a person having to learn Another Occupation. Or do you think programmers, good ones, just know Programming?

Some folks do combine excellent programming skills with excellent business skills, but that's not necessarily what everyone should aim for.

I find that you can score someone on a whole range of skills, much as you can on computer games. The more overall skills you have, the more you will cost. If a job requires excellent programming skills but doesn't need any business skills, then paying for someone that has excellent business sills would probably be a waste of money.

In terms of your parent point about whether all business analysts should learn some programming, I think that programming skills are definitely helpful for business analysts, but they are neither necessary nor sufficient. Different skills help in different situations.

I'm curious, is there a way to "profile" an excel application? If you suspect a sheet is taking longer to calculate than it should are there tools that can help you drill down to discover the bottleneck?

Great question; I'd love to read an answer.

I'm curious if that analysis could feed a parallelization step. Then you could offload work chunks to a compute farm.

I don't think you even need to go that far. Can you optimise your implementation so that parallelisable chunks could run across all your cores? This sounds like the kind of think Excel may even do "automatically" but like many such things you usually need to structure your implementation in a suitable fashion.

It would be kind sad if MS didn't already do that.

It's a NP complete problem. What I mean is, there is no generalisable solution to this. Optimising algorithms employed by sophisticated technical users (think `-O`optimisation; Java hotspot) typically employ sets of heuristics that run the code in more optimal ways based on certain patterns. To trigger these you usually have to be aware of them and write your code in a certain way.

Given that Excel isn't really targeted at these kinds of user (heavy lifting typically offloaded to external libraries) it wouldn't be "Sad" if they didn't already do it, and I would in fact be pleasantly surprised. Which is why I'm asking the question.

EDIT: Turns out you're right! https://msdn.microsoft.com/en-us/library/office/bb687899.asp...

  To trigger these you usually have to be aware of them and write your code in a certain way.

Compiler writers also look for common patterns in people's code and figure out how to optimize them. when you have both years (decades) of legacy code and developers who don't even know what patterns are optimized, you as a compiler writer need to optimize the code that is being written.

Yes that's what I meant. Common patterns = Heuristics as a developer you need to be aware of the common patterns the compiler is looking for.

Excel uses multiple cores for calculations. You can configure how many to prevent it sucking up the entire CPU for long calculations.

It's a pretty easily parallelizable problem since the sheet is just a dependency tree with the cells as nodes.

The build-in functions (e.g. probability distributions etc.) could also be multi-threaded, although I'm not sure if they are. Our external API called from cells was written in C++ and already multi-threaded.

More info here:

https://msdn.microsoft.com/en-us/library/office/bb687899.asp...

ex Excel PM. There is a project that was between the Excel team and the high performance computing team at Microsoft for exactly this purpose. Not surprisingly mostly used by investment banks and insurance companies. (https://msdn.microsoft.com/en-us/library/ff877825(v=ws.10).a...)

I googled it: "Excel Profiler" turns up a number of 3p products such as this http://www.decisionmodels.com/FastExcelV3Profiler.htm

On decisionmodels, the website looks like it is coming straight from the 90s (but after all so does excel) but it is a gold mine of stuff to know before one can claims to really understand excel. I highly recommend the reading to anyone aspiring to be a "poweruser".

Seconded. decisionmodels.com has excellent stuff on the difference between F9, sh-F9 and ctrl-sh-F9, and explains when the formula graph gets rebuilt. When you're building calc heavy XLL addins it's important to have a clear understanding of the behaviour of the calc engine code invoking your addin.

Reply on news.ycombinator.com