[求助] Ibank面试真题 (final round)

 Testing a simple investment strategy on S&500
The S&500 is an index based on 500 US stocks, many of which are the largest in the US markets. Information:
• We have daily open/high/low/close data of the 500 stocks from 1/1/2000
• We have a formula for an indicator, which gives out a value between [0,100] for each stock based on their historical prices

Supposedly, the indicator tells us whether a stock is a good buy/sell – higher the value of the indicator (closer to 100), the better the stock performs, and vice versa.
With the data in hand, we want to figure out if this indicator works on the stocks in the S&500 index.

Key Discussion Points:
1. How to approach the testing and why? It is ok to assume monthly rebalancing of stock portfolio.

2. We have slightly more than 9 years of data history. Assume that, based on your testing results, the indicator works or doesn’t work. Will this performance trend continue in 2009 and beyond? How confident can you be about the future performance of this strategy?

3. How would you actually do the backtesting? [This is a technical question regarding computer programming and data manipulation – interested in how you would actually do the backtesting given the above data]  (Can use SQL and R program)

请求大牛解救？

有没有IT部的牛人指导一下??

顺便提示一下IT部的前途如何. 谢了

都是很基本的东西吧？

1 重新计算过去9年的实际股票回报率（包括分红，分股）

2 用你这个公式，回头计算过去9年股票的得分

3 对两者作线行回归/统计分析。。得出mean, median, R^2, std dev, correlation........

两个方法，你的测试方法可以取这两种方法的交集。

1.观察指标在标准普尔500指数出现拐点时候的指标数值，如果每次拐点指标数值都是有效的，或者明确地预测出了市场拐点，那么指标是可行的。反之，如果在市场出现拐点或者将出现拐点的时候，指标出现钝化或者不能预测出指数将来走势，那么该指标无效。

2.用问题3给的提示，即用SQL数据库按指标发出的指数，即在历史数据中按这个指标发出的信号进行股票买卖测试操作，在指标数值显示该买入的时候买入，该卖出的时候卖出。用这几年的历史数据进行测试，如果测试出来的收益率高于同期标普500涨幅，那么说明指标还凑活（英文表达自己解决）。

Will this performance trend continue in 2009 and beyond? How confident can you be about the future performance of this strategy?

No, 09年金融危机下，市场充满了非理性。如此情况下，市场是无效的，不能反映证券资产的价值。没有人能预测非理性的程度，指标也不例外

这种东西细节才是最主要的，比如说你得分高还是低 buy/sell这个预期时间段是多长？1m?3m?6m?比如如果你得出的结果是：得分和实际回报有很高的correlation,但同时在数据分布的两端有很大的误差, 你该如何解读？比如说如果得分和实际回报有误差的话，如何具体分析这个误差如何分析它是否有某些bias。。诸如此类。。。

以我的工作经验看，如果你真的能思路清楚，很多细节都有分析到，过个final round应该足够了。一个面试而已，又要你直接写SQL,能有多复杂？考考你对金融的了解，对统计的了解，对SQL等具体技术的掌握，再看看你有没有能力把3者联系到一起，这也就是了。。