Discopolis - Timber Merchants
http://soundcloud.com/discopolismusic
Music video for Scottish band Discopolis' song 'Timber Merchants'.
FIND US ON - Builders Merchant, Builders Supplies, Home ...
Ronseal Decking Stain 2 .5l £ 19.20 : inc VAT : £16 .00 +VAT each: Country : Oak Mountain : Green Rich Teak: Golden Cedar Rich Mahogany: Rustic Pine: TIMBER MERCHANT TIMBER MERCHANT: 15 YEAR DESIRED : SERVICE LIFE 15 YEAR DESIRED : SERVICE LIFE: WHETHER YOU WANT TO CREATE A ...Quantile Forecasting Technology Part I
In order to understand quantile forecasting, we need to take a brief look at the general practice of inventory optimization and why we forecast in the first place.
Inventory optimization is the task of finding the level of inventory that balances the cost of excess inventory with the cost of stock outs. In practice, we do this by making two decisions:
• When to order... and....
• How much to order
The first is given by the reorder point, which is the minimum stock on hand that should trigger a backorder. The second is given by the optimal order quantity, which is the number of units that should be ordered.
Let's focus on the reorder point and how it is determined using classical forecasting methods. The reorder point is the inventory quantity that should trigger a backorder. Intuitively, the reorder point is the quantity of inventory that covers continuing sales while waiting for a backorder to arrive (called lead demand), plus the unpredictability or volatility in demand (called safety stock). Both lead demand and safety stock require a forecast.
The type of forecast that is used today for inventory optimization is the classical 'mean forecast', which is produced by using averages, exponential smoothing, Holt Winters, and other 'classical' forecasting methods. These forecasts are called 'mean forecasts', because the probability of over- and under- forecasting are strictly balanced. In other words, following the recommendation rendered by the forecast will give you equal chances of having a stock out and having too much inventory for actual demand. The implicit service level, which is the probability of not experiencing a stock-out, is 50 percent.
This scenario would not be a problem if the cost of over and under forecasting was equal. However, for retail companies as for many other businesses, this is not the case. Allocating too little resources 50 percent of the time is a poor trade-off that does not reflect the reality of the business. In practice, the marginal cost of a stock-out vastly exceeds the marginal cost of additional inventory. This is why service levels in retail are typically chosen as being above 90 percent, which means that 9 out of 10 orders can be serviced from stock on hand, and in only 1 out of 10 orders will a stock out occur. As a concrete example, food retailers choose their service level typically between 95 and 98 percent, which reflects the much higher cost of a stock out versus additional inventory.
This business-specific asymmetry is the reason that companies purposefully introduce what is called a 'bias' in their resource allocations. This is done by adding safety stock to increase the service level from the 50 percent provided by the mean forecast to the desired level (in our retail example above 90 percent). Mathematically, it means nothing else but producing a quantile forecast via extrapolation, or an 'extrapolated quantile forecast'.
The problem with this methodology is a negative effect on accuracy, which is particularly poor in three cases:
• high service levels
• infrequent sales of an item
• and spiky demand.
Unfortunately, one or several of these cases are the norm for retail, wholesale and manufacturing businesses.
'Native' quantile forecasts represent a radical improvement from classical forecasts in terms of accuracy. Benchmarks against our classical forecasting engine have shown an increase in forecasting accuracy of between 20 percent to over 50 percent for businesses offering car parts, electrical supplies, spare parts, textiles, luxury goods, chemicals, and groceries.
The key idea of quantile forecasting is that the bias is introduced on purpose from the start -in order to alter the odds of over and under forecasting. Instead of producing a mean forecast, or 50 percent quantile, that is then extrapolated to reach the desired service level while suffering a loss in accuracy, the forecasting model directly produces a quantile forecast by taking into account the desired service level. By doing so, it consolidates the several steps involved in calculating a reorder point in the classical way, at the benefit of a higher accuracy. Quantile forecasts are operated and refreshed in a similar way and at the same frequency as mean forecasts.
Quantile forecasts are applicable to all inventory-related forecasting needs. In our opinion, they will make classical forecasting largely obsolete in inventory optimization within the next 10 years.
View Doc