6th Dec

The snippet for ACF and PACF is:

PACF :

ACF:

For ‘hotel_avg_daily_rate,’ the PACF indicates a notable spike at lag 1, followed by subsequent lags within the confidence range, suggesting an AR(1) element in the SARIMA model. The decreasing trend in the ACF suggests the need for differencing (d) to achieve stationarity and potentially a non-seasonal MA component.

On the other hand, ‘hotel_occup_rate’ exhibits a prominent initial spike in the PACF and significant seasonal spikes in the ACF, hinting at possible seasonal MA components. This points toward a SARIMA model with an AR(1) part and seasonal differencing, likely SARIMA(1,1,0)x(0,1,Q)12. Here, ‘Q’ relates to significant seasonal lags observed in the ACF plot. Determining the exact ‘Q’ value requires further scrutiny of these seasonal lags, yet the evident seasonality implies it wouldn’t be zero.

 

4th Dec

In this update, we’re delving deeper into the SARIMA model’s application in analyzing the BPDA economic indicators dataset. Understanding the connections between Autocorrelation Function (ACF), Partial Autocorrelation Function (PACF), and SARIMA is crucial for leveraging this model effectively. The ACF indicates how a time series correlates with its past values at different time lags, while the PACF reveals the isolated impact of each lag on the series, separate from shorter lags. Determining the SARIMA’s AR terms (p) relies on significant spikes in the PACF plot, signifying potential Autoregressive elements. Meanwhile, the ACF plot aids in identifying the suitable number of Moving Average terms (q) based on significant spikes at specific lags. Similar principles apply to seasonal aspects (P, D, Q, s) in SARIMA. Seasonal spikes in these plots help determine the seasonal components, with their positions guiding the selection of seasonal periods. Utilizing ACF and PACF analyses on columns like “hotel_occup_rate” and “avg_daily_rate” allows a better grasp of their inherent patterns, steering us toward crafting SARIMA models. For example, if the ACF plot of ‘hotel_occup_rate’ shows a clear seasonal pattern and the PACF plot demonstrates an abrupt cut-off, it suggests integrating seasonal AR and non-seasonal MA components into our SARIMA model.

The subsequent update will detail the functionality of Python code for ACF and PACF, demonstrating its integration with the SARIMA model.

1st Dec

Forecasting involves using statistical models to predict future values of a variable, such as sales or economic indicators, by analyzing historical patterns. Its importance lies in aiding planning and decision-making, guiding strategies, budgeting resources, and managing inventory.

Forecasting finds application across various sectors: governments use it for policy planning, retail for inventory management, transportation for optimizing schedules, and finance for predicting market movements.

The process typically involves data collection, analysis, model selection, and generating forecasts. The choice of model depends on the data and goals, ranging from simple trend extensions to complex computer-based predictions. In our project, one method employed is SARIMA, or Seasonal AutoRegressive Integrated Moving Average, specifically suited for time series data with seasonal patterns like economic indicators.

29th Nov

In my latest update, I’ll continue from where I stopped before. I’ll show how I used Python to include the ADF and KPSS tests in my project. I’ll also talk about what I learned from the results of these tests and what they mean for my analysis. Here’s a piece of the code I used in Python

 

27th Nov

Stationarity matters because it affects how good our predictions about the future can be. When data isn’t stationary, our predictions might not be very accurate. To check if data is stationary or not, we use two tests: the ADF test and the KPSS test.

The ADF test helps us figure out if the data is changing too much over time. If the result is smaller than a certain number, it means the data is probably okay for making predictions. On the other hand, the KPSS test checks if there’s a pattern or trend in the data. If the result is higher than a specific number, it means there might be a trend that could mess up our predictions.

We also use something called the Autocorrelation Function (ACF) alongside these tests. ACF helps us see how data points relate to each other over time. This helps us pick the right tools to make predictions, especially for models that rely on understanding how data behaved in the past to guess what might happen next.

24th Nov

I present to you the main conclusions of our Time Series Analysis of the BPDA economic indicators dataset. Upon analyzing the data using statistical approaches, a few noteworthy patterns and trends have surfaced. We’ve included graphic representations of the patterns we’ve seen with this release.